Evaluative Degree Modification
of Adjectives and Nouns
Hanna de Vries
MA Thesis - Linguistics: the Study of the Language Faculty
Utrecht institute of Linguistics OTS
Supervisor: dr. Rick Nouwen
Second reader: dr. Yoad Winter
Abstract
It is a well-known but little-studied fact that evaluative adverbs - adverbs indicating the attitude of the speaker towards the information she is conveying can modify degree (incredibly tall, ridiculously expensive...). This thesis offers
a syntactosemantic account of evaluative degree modification of both gradable
adjectives and gradable nouns.
Following Morzycki (2004), I propose that evaluative degree modification
involves a covert operator (which I will call eval); however, my proposal differs
from that of Morzycki in several crucial respects. Most importantly, I argue
that evaluative degree constructions should not be analysed as embedded exclamatives. Furthermore, I show how their syntactic behaviour illuminates their
semantic composition, using evidence from different phenomena in both English
and Dutch.
Subsequently, I examine the linguistic evidence for the gradability of certain
nouns, like idiot, nerd, genius, Barbie doll enthusiast, and weirdo, and conclude
that they, like gradable adjectives, have a degree argument. I show how this
class of gradable nouns may be defined in prototype-theoretic terms. Morzycki
(2009) has shown that gradable nouns can be modified by size adjectives like
big and enormous; I extend his account by including degree modification by
evaluative adjectives.
Finally, several suggestions for further research are offered.
1
Acknowledgements
First of all, I want to thank Rick Nouwen for being a great supervisor, from the
moment I first walked into his office with zero knowledge of anything degreerelated, to where I am now, hoping to have provided a meaningful addition to
the field. Rick gave me all the time, space and independence I needed to develop
my own thoughts on where this thesis should go. Also, while never short on
constructive criticism, he gave me the feeling that my linguistic thoughts are
worth writing down, even if they are not about to revolutionise the field. (Yet.)
I also feel very honoured that Yoad Winter agreed to be second reader of this
thesis. During the past two years, almost any subject I took up for some paper
or other (including this thesis) turned out to be something Yoad had already
worked on, so I have come to regard him as some kind of semantic Superman.
The following people provided the data in Chapter 3: Marko Hladnik (Slovene),
Anne Morès (French), Liquan Liu (Mandarin), Misako Tanaka (Japanese), Joe
Wolfson (Hebrew), and Linguaphiles amles80 (Swedish) and virre (Norwegian).
Data-wise, I also owe a lot to the Linguaphiles community at Livejournal.com
(communities.livejournal.com/linguaphiles), whose native English speakers never
got tired of providing judgements of every crazy sentence I came up with, kindly
pointing out to me that whatever I wanted to say could be expressed much more
clearly without all these awkward degree constructions, and bravely suggesting
third options when I asked them to pick the best alternative from two possible
sentences - if I ever appeared impatient or less thankful than I should have been,
I’m very sorry.
My final thanks go to my wonderful flatmates (in particular Arthur and Maarten),
for supplying me with coffee throughout the writing process and even occasionally feigning an interest in whatever was keeping me so busy; my parents for
supporting me throughout the past seven years (and before), even though I
chose a field of study that will get me neither rich, nor famous, nor particularly
useful to society (or even guarantee me a job); and last but not least, Matthijs
Westera for some proofreading, lots of mental support, and more.
2
Contents
1 Introduction
1.1 Gradability and degree . . . . . . . . . . . . . . .
1.1.1 The semantics of gradable predicates . . .
1.1.2 The Degree Phrase . . . . . . . . . . . . .
1.1.3 Covert degree morphology: pos and meas
1.1.4 Properties of the degree scale . . . . . . .
1.2 The semantics of degree modifiers: some issues .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
7
7
9
12
16
20
2 Evaluative adverbs as degree modifiers
23
2.1 Previous accounts: Morzycki (2004) . . . . . . . . . . . . . . . . 23
2.2 Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.1 Some arguments against EDCs as embedded exclamatives 27
2.2.2 In which monotonicity saves the day . . . . . . . . . . . . 29
2.3 Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.1 Not in SpecDegP, but in SpecQP . . . . . . . . . . . . . . 31
2.3.2 r as a covert Q-head: evidence from much-support . . . . 33
2.3.3 r as a covert Q-head: evidence from Dutch gender agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Assembling the pieces . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.1 Splitting up r . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.2 Back to the denotation of gradable adjectives . . . . . . . 38
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 Degree modif ication of gradable nouns
3.1 What makes a noun gradable? . . . . . . . . . . . .
3.1.1 Linguistic versus conceptual gradability . . .
3.1.2 Monotonicity . . . . . . . . . . . . . . . . . .
3.1.3 Scales and prototypes . . . . . . . . . . . . .
3.2 Syntax and semantics of noun modification . . . . .
3.2.1 Degree modification by size adjectives . . . .
3.2.2 Degree modification by evaluative adjectives
3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . .
3
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
43
45
46
50
50
56
60
4 Suggestions and further issues
4.1 Evaluative modifiers across languages
4.1.1 Data . . . . . . . . . . . . . . .
4.1.2 Some tentative conclusions . .
4.2 Gradable verbs . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
61
61
62
64
66
5 Conclusions
70
References
72
Appendices
77
4
Chapter 1
Introduction
Although many researchers have noted that gradability is not just a property of
adjectives (cf. Bolinger 1972, Abney 1987, Kennedy & McNally 2005), or that
certain degree expressions (like French trop) are able to combine with a whole
range of lexical classes (cf. Doetjes 1997, 2008), there are few syntactic and
semantic theories of gradability in non-adjectives. Of course, not all apparent
gradability, or compatibility with certain modifiers, may stem from the same
source; for example, while more (and less) can be used to modify all expressions
below, there seems to be an intuitive difference between what is happening in,
say, (1a) and (1d-e):
(1)
a. Vernon is more intelligent than Maxwell.
b. You need to exercise more.
c. Maxwell appreciates jazz more than Vernon.
d. This party needs more beer.
e. This sentence contains more words than the previous one.
One way to try and put this difference into words is that the former involves
degree - we can talk about lower and higher degrees of intelligence, for example,
but not about degrees of beer, or degrees of words. Beer and words can occur in
lower or higher quantities, but even so, they are either there - fully, undeniably
- or they are not. It is not so simple with a property like intelligence (or height,
or weight, or age...). I can state that Maxwell is not intelligent, but even a
low degree of intelligence is a degree of intelligence, so in a way the property
denoted by intelligent still applies to Maxwell. The same might hold for a
verb like appreciate. I can say that Vernon appreciates jazz less than Maxwell,
without implying that either of them actually likes jazz - maybe Maxwell just
dislikes it less than Vernon.
So, to delimit our subject, we could centralise this notion of degree and state
that the kind of gradability we want to talk about is present in the lexical semantics of the gradable element in the form of a degree argument. Thus, an
individual cannot just have the property of intelligence - this is semantically
5
incomplete. We need to know to what degree this property holds of said individual; a gradable predicate cannot be interpreted if its degree argument is not
saturated.
This means that we do not have to talk about quantification over event
verbs (like exercise) or count and mass nouns1 . Leaving experiencer verbs like
appreciate aside (for now - I will return to the issue in Chapter 3), it does seem
that gradability in this narrow sense primarily concerns adjectives.
However, there is a class of count nouns that behaves exactly like gradable
adjectives (modulo independent syntactic differences). Its most clearly defineable members are nouns like idiot, nerd, fan, psychopath, airhead, enthusiast,
simpleton, loser and weirdo - nouns that denote a person who possesses a certain gradable property. In Dutch, one can form them straightforwardly from
gradable adjectives (at least those of Germanic origin) by means of the derivational suffix -erd : slimmerd from slim ‘smart’, dikkerd from dik ‘fat’, gemenerd
from gemeen ‘mean’, and so on. Recently, Morzycki (2009) developed a theory
of the syntax and semantics of the modification of such gradable nouns by size
adjectives like big:
(2)
a. George is an enormous idiot.
b. Three huge goat cheese enthusiasts were arguing in the corner.
c. George is a bigger idiot than Floyd.
Although all three examples have readings in which the size adjective refers to
the physical size of George or the others, this is not the most natural reading.
Rather, we tend to interpret these size adjectives as claims about the extent
of George’s idiocy or the goat cheese lovers’ enthusiasm: (2a) thus means that
George is extremely idiotic, and so on.
But size adjectives are not the only adjectives that can modify gradable
nouns. Many adjectival counterparts of so-called evaluative adverbs share with
their adverbial sisters the possibility of modifying degree:
(3)
a. Maxwell is incredibly stupid.
b. Maxwell is an incredible idiot.
(4)
a. Vernon is unimaginably tall.
b. Vernon is an unimaginable goat cheese enthusiast.
In this thesis, I will focus on evaluative degree modification of adjectives as
well as nouns, reviewing and extending two papers by Morzycki: one on evaluative modification of adjectives (Chapter 2), and the aforementioned account of
noun modification by size adjectives (Chapter 3). The remainder of this first
chapter will be devoted to an introduction on gradability and degree in general.
1 Nicolas (2004) shows that the notion of degree is in fact relevant to mass nouns derived
from gradable adjectives (like wisdom and generosity). In this respect, they behave differently
from ‘ordinary’ mass nouns (like beer and furniture); for example, like gradable adjectives and
count nouns, they can be modified by a size adjective (enormous wisdom).
6
Subsequently, in Chapter 2, I will develop a new analysis of evaluative degree
modification, which I will extend to gradable nouns in Chapter 3. Chapter 3
further contains a proposed definition of what exactly constitutes gradability
in nominal predicates, and a discussion of some remaining issues. Chapter 4
contains the conclusion.
1.1
Gradability and degree
The literature on gradability and degree can be hard to get into, as there exist many different analyses even of the most basic elements involved, and researchers are not always explicit in their assumptions (let alone in their rejection
of the alternatives). This section contains my attempts at distilling from the
existing literature a set of basic assumptions on which to found the rest of this
thesis, including their history and some possible alternatives. In section 1.1.1 I
briefly discuss the semantics of gradable adjectives. Section 1.1.2 introduces the
Degree Phrase, which provides the syntactic basis for this thesis. Section 1.1.3
introduces covert degree morphology. Finally, sections 1.1.4 and 1.2 discuss degree modification by words like very and completely, and how this relates to
certain properties of the gradable adjective.
1.1.1
The semantics of gradable predicates
Intuitively, the denotation of a degree construction like six feet tall should resemble something like the following:
(5)
[[six feet tall]]he,ti = λx∃d[tall(d)(x) ∧ d = 600 0]
This denotation has several components. There is a degree predicate, tall,
that relates an individual x and x’s height d such that tall(d)(x) means that
x is tall to the degree d; furthermore, d is compared to (in this case, equated
with) a particular value, in this case six feet. So, the meaning of a sentence
like Vernon is six feet tall can be paraphrased as ‘There is a degree d such
that Vernon is d-tall and d equals six feet’. Similarly, the semantics of Vernon
is taller than Maxwell should consist a degree predicate relating Vernon to his
height, and a comparison between that height and some other value (provided,
in this particular case, by Maxwell’s height). The comparison function here is
not equality, but a greater-than relationship; had we used less tall or as tall
as, it would have been yet another different function. This indicates that the
comparison function is provided by elements like less, more, -er, as...as, and too,
the so-called ‘degree morphology’. The comparison value, as we saw, is provided
by elements like six feet or than Maxwell. The gradable adjective provides
the degree predicate, which means we will assume the following denotation for
gradable adjectives:
(6)
[[AP]]hd,he,tii = λdλx[G(d)(x)]2
2 As
mentioned before, degree predicates relate individuals and degrees. This does not
7
While the above ingredients - degree predicate, comparison function, comparison value - all need to be present in the semantics, they may be absent from
overt syntax. The so-called ‘positive form’ (Vernon is tall ) intuitively involves
Vernon’s height being favourably compared to some contextually defined standard degree, but neither the comparison nor the standard are overt. Similarly,
Vernon is six feet tall lacks an overtly stated equality relationship between Vernon’s height and the degree of six feet, yet it is obviously there in the semantics.
We solve this by assuming covert degree morphology, which I will discuss in
more detail in section 1.1.2.
In the above approach, degree constructions (like six feet tall ) are analysed
in terms of existential quantification over the degree argument d. However,
Kennedy (1997) argues that this approach leads to wrong predictions about
the scopal behaviour of degree constructions. If degree constructions involved
quantification, we would expect them to show scope ambiguities in sentences
like (7a), just like other quantificational expressions (the indefinite DP in (7b)).
(7)
a. Vernon isn’t taller than Maxwell is.
b. Vernon didn’t see a black cat.
In (7b), a black cat can have either wide scope (‘there is a black cat such
that Vernon didn’t see it’) or narrow scope (‘there is no black cat such that
Vernon saw it’). Similarly, we might expect (7a) to be ambiguous between
a narrow scope (‘there is no degree d such that d exceeds Maxwell’s height
and Vernon is d-tall’) and a wide scope interpretation (‘there is a degree d
such that d exceeds Maxwell’s height and Vernon is not d-tall’). The latter
interpretation is compatible with there being some other degree d0 such that d0
exceeds Maxwell’s height and Vernon is d0 -tall - which, obviously, contradicts
our natural understanding of the meaning of (7a). In other words, degree heads
do not seem to behave like quantifiers in this respect (but see Heim (2000) for
some counterevidence).
This leads Kennedy to reject the degree predicate treatment of gradable adjectives and reduce them to measure functions, i.e. functions from individuals to
degrees (type he, di). In effect, this treats gradable adjectives like a scale, along
which individuals are measured. The adjective does not have a degree argument; rather, it takes an individual and returns a degree. In this approach, the
denotation of tall, for example, is λx[height(x)]; degree morphology functions
to establish a comparison relation between height(x) and some comparison
value.
However, the choice between he, di and hd, he, tii does not seem a very principled one. Often, the type adopted depends on the particular goals of the work
in question; for example, while Kennedy (1997) en Svenonius & Kennedy (2006)
necessarily mean that every individual can be related to only a single degree. In fact, we will
assume that individuals are related to a whole range of degrees; thus, for an individual x with
a height of six feet, tall(d)(x) is true for d=600 0, but also for all degrees d with a value lower
than 600 0. More on this in chapter 2.
8
opt for he, di, Kennedy & McNally (2005) assume hd, he, tii. Similarly, Morzycki’s (2004) account of degree modification by evaluative adverbs (remarkably,
surprisingly) assumes a degree predicate approach, while his analysis of gradable nouns (Morzycki 2009) is measure phrase-based. It must also be noted that
the hd, he, tii type is compatible with several slightly different denotations, so
some additional semantic variety exists.
Throughout this thesis, I will assume that gradable adjectives denote degree
predicates (type hd, he, tii, denotation as in (6)). This is not just for the sake
of clarity and exposition: as we will see, it will become crucial in Chapter 2.
However, as the measure phrase approach will make a brief return in Chapter
3, the reader is asked to bear its existence in mind.3
1.1.2
The Degree Phrase
In the earlier days of generative syntax, degree words were taken to occupy the
specifier position of the adjective phrase (e.g. Jackendoff 1977:8), as in (8):
(8)
AP
A0
DegP
A0
Compl
This corresponded to the general view of syntactic structures being projected
by lexical heads. However, following the work of Chomsky (1986), Fukui &
Speas (1986) and Abney (1987), among others, more and more phrase structures
were taken to have a functional rather than a lexical head - most famously,
S0 became IP, S became CP (Chomsky 1986), and NP became DP (Abney
1987). Similarly, a functionally-headed alternative for the traditional AP was
also proposed (Abney 1987, Corver 1991, 1997a,b): DegP, the degree phrase.
3 Although
I have only discussed approaches that involve degrees (elements of type d), not
all analyses of gradability in fact do. The main alternative to degree-based approaches (for an
overview of its proponents see Kennedy 1997, Rett 2008, van Rooij (to appear)) is to analyse
gradable adjectives as ‘vague’. One way to do this is to assume that gradable adjectives,
like non-gradable ones, denote functions of type he, ti, but this is a partial function: for
some individuals x, the truth value of Adj(x) remains undefined. Whether Adj(x) is true,
false or undefined is determined by interpreting it relative to some comparison class, which
is context-dependent (Klein 1980). For arguments against this view, see von Stechow (1984)
and Kennedy (1997). See also Van Rooij (to appear) for a recent discussion of various views
on gradability and vagueness.
9
(9)
DegP
Deg0
Spec
Deg0
QP
Q0
Spec
Q0
AP
Originally, AP was taken to be the complement of the degree word in Deg0 , and
degree modifiers were located in SpecAP; the additional Quantifier Phrase level
was argued for by Corver (1997a,b), following similar suggestions by Bresnan
(1973) and Creswell (1976). In this so-called ‘split DegP’, not all degree morphology occupies the same position: Deg0 contains degree morphemes like so,
how, too, and as, while Q0 contains the quantifiers less, more/-er, and enough.
SpecDegP can be occupied by various elements, in particular measure phrases.
Modifiers like very and extremely appear in SpecQP.
Before AP became DegP, there was no structural way to distinguish ‘degree
words’ from ‘modifiers’. Thus, very tall and too tall had precisely the same
syntax, with both very and too located in the Spec of AP. One important empirical reason to adopt the DegP in (9) is that degree words and modifiers,
in fact, do not behave identically at all. Corver (1991, 1997b) argues extensively for the claim that they occupy different syntactic positions. For example,
they can appear together, as in (10), and show different behaviour in several
constructions (11-12) (all examples from Corver (1991):
(10)
a. How very interesting!
b. Zo heel knap is ze anders niet.
(Dutch)
that very pretty is she however not
‘She’s not that very pretty at all’
(11)
a. Too/how/that big a car
b. *Very/extremely big a car
(12)
a. Though the house is very expensive...
b. Very expensive though the house is...
c. Though Mary is too shy for her own good...
d. *Too shy for her own good though Mary is...
Furthermore, the fact that gradable modifiers can themselves be modified - i.e.
form modifying phrases, of potentially infinite length - shows that they cannot
be heads (cf. Morzycki 2004):
(13)
a. [pretty [insanely]] crazy
10
b. [very [remarkably]] tall
c. [fairly [embarrassingly]] angry
If we were to maintain that modifiers are heads and propose an alternative
structure like [pretty [insanely crazy]], this would force us to assume that degree
phrases can take other degree phrases as their complement. Obviously, this leads
to the wrong predictions (for example, sentences like *Vernon is too six feet tall
or *How six feet tall is Vernon? should be fine under this assumption). But
more importantly, it is directly at odds with the reason that degree phrases exist
in the first place - namely, to deal with gradability. Degree phrases themselves
can, by definition, not be gradable; therefore, they cannot be selected by a
degree head, which, by definition, needs a gradable complement.
Alternatively, one might suggest that the adverbs here do not modify each
other, but rather all modify the adjective, intersectively - just as in big expensive
red car the adjectives do not modify each other but the noun. There are a
few objections to this. First, the intonation patterns differ: if the modifiers
are all on the same level, modifying the same element (as in big expensive red
car or extremely, insanely, embarrassingly angry), they all bear intonational
stress. However, if they modify each other (as in extremely red car or fairly
embarrassingly angry), only the first modifier bears stress. That the latter
stress pattern is grammatical in (13) indicates that the structure is indeed as
indicated.
Secondly, if the adverbs were all equally modifying the adjective, we would
expect that both gradable and non-gradable modifiers could be preceded by
others. However, this expectation is not borne out: *insanely pretty crazy,
*remarkably very tall, and *embarrassingly fairly angry are all ungrammatical.
While this could, in theory, be due to some independent ordering criterion
(comparable to the one that rules out *?red big car, for example), it is extremely
suggestive (to say the least) that the effects of this criterion just happen to
coincide with the independent predictions of the ‘modified modifier’ analysis.
It should be obvious that the latter analysis is much more parsimonious than
the former.
The above shows that there are clear structural differences between degree
heads and degree modifiers. So, if modifiers occupy a specifier position, which
one is it? In principle, we have three options: SpecDegP, SpecQP and SpecAP.
The latter is ruled out, as evidenced by the fact that we can replace the AP
with the anaphor so and still have a modifier (Vernon is tall, even remarkably
so). The former predicts that the modifier will precede the degree word in
sentences that have both, which is at odds with (10). This points to SpecQP
as the position for the latter. (For a much more detailed argument, see Corver
(1997a,b).) I will deal with the semantic side of modifiers in section 1.2.
In the remainder of this chapter, I will mostly use a slightly simplified version of DegP, as the distinction between DegP and QP will not be semantically
relevant until halfway Chapter 2.
11
1.1.3
Covert degree morphology: pos and meas
Looking at the data in (14), it may seem that an adjective like tall is ambiguous
between a property of type he, ti - in other words, behaves like a non-gradable
adjective - and a degree predicate of type hd, he, tii.
(14)
a. Vernon is tall.
b. Vernon is quite tall.
c. Vernon is five feet tall.
In (14a), the so-called ‘positive form’, tall denotes the property of having a degree of height that exceeds some contextually determined standard (with respect
to some comparison class, say adult men if Vernon is an adult man), whereas
in (14b-c) it denotes a degree predicate, the degree of height being indicated by
quite and f ive feet, respectively. Furthermore, (14c) does not entail (14a) (in
this case, (14a) is false if (14c) is true). All in all, the tall in (14a) seems to be
of a rather different kind than the tall in (14c).
But it is not generally assumed that gradable adjectives are ambiguous.
Rather, the general approach is to assume a covert degree morpheme in Deg,
which is applied to the adjective to yield the property reading. This morpheme,
pos, was (to my knowledge) originally proposed in Cresswell (1976) (and see
further von Stechow 1984, Kennedy 1997)4 . (15) illustrates how it works:
(15)
DegP
Deg0 he,ti
Spec
Deghhd,eti,he,tii
APhd,he,tii
pos
tall
Assuming pos ensures a taller-than-d0 reading, where d0 is some contextually
defined standard or average, we may wonder what happens when SpecDegP
contains a measure phrase, as in (14c). This is potentially problematic, as
Zamparelli (1993) notes. Independently from the existing semantic literature
on pos, he also argues for the presence of a null degree head, which he calls
OP. Semantically, it does precisely the above: ensuring that the positive form
gets a taller-than-standard interpretation. But in order to have this analysis
account for the contrast between (14b) and (14c) - unlike (c),(b) does entail
(a) - Zamparelli is forced to stipulate that “the semantic effect of OP in Deg is
cancelled when there is some element in [SpecDegP]”. After all, we assume that
measure phrases like f ive feet occupy the SpecDegP position, while modifiers
4 Kennedy & McNally (2005) note that according to Sybesma (1999), Mandarin marks the
positive form overtly, with the degree morpheme hen. This might be taken as crosslinguistic
evidence for the existence of pos. See Liu (2009) for more on hen and Chinese degree phrases
in general.
12
like quite do not (they appear in SpecQP). Obviously, this is not a very satisfying
solution, as it results in a non-compositional degree semantics.5
Kennedy (1997), the first to base an extensive semantic account of gradability and degree constructions on Abney’s and Corver’s Extended AP, proposes
a semantics for pos that does not result in problems with compositionality.
In Kennedy’s analysis, all degree morphemes have the same basic denotation,
(16a), which makes (16b) the denotation of pos:6
(16)
a. [[Deg]] = λGλd0 λx∃d[G(d)(x) ∧ R(d)(d0 )]
b. [[pos]] = λGλd0 λx∃d[G(d)(x) ∧ pos(d)(d0 )]
c. [[pos(d)(d0 )]] = 1 iff [[d]] ≥ [[d0 ]]
Here, G is the denotation of some gradable adjective, d0 is the comparison
value, and R is a relation introduced by the particular degree morpheme. In
the case of pos, this is a partial ordering relation (≥) between the actual and
the comparison value.
Crucially, the comparison value is not part of the denotation of pos itself,
but has to come from elsewhere. If DegP contains a measure phrase in its Spec,
this will determine the standard value. If it doesn’t - as with the positive form a variable (here sG , which stands for ‘the standard of G’) is introduced at DegP
level:7
(17)
a. Vernon is tall: ∃d[tall(d)(v) ∧ pos(d)(stall )]
5 Klein
(1980) rejects Cresswell’s pos-morpheme for similar reasons.
Kennedy’s original analysis, gradable adjectives denote measure functions; I have
modified it to fit in with the degree predicate approach outlined above. (A more explicit
example of such a translation can be found in Chapter 3, section 3.2.1.) Also, Kennedy
himself uses abs (for ‘absolute’) rather than pos.
7 Tree adapted (with a small correction) from Kennedy (1997)
6 In
13
b.
IP
λx∃d[tall(d)(x) ∧ pos(d)(stall )](v)
DP
Vernon
VP
λx∃d[tall(d)(x) ∧ pos(d)(stall )]
V
is
DegP
λd0 λx∃d[tall(d)(x) ∧ pos(d)(d0 )](stall )
Deg0
λGλd λx∃d[G(d)(x) ∧ pos(d)(d0 )]
(λdλx[tall(d)(x)])
0
Deg
AP
λGλd0 λx∃d[G(d)(x) ∧ pos(d)(d0 )]
tall
But as Kennedy himself already notes, this account has its problems as well. If
the standard value enters the derivation by means of an indexical variable, it
should remain constant under VP ellipsis, as (18a) shows (that must refer to
the same drink in both conjuncts). This means that (18b) can only be true if
father and daughter are tall with respect to the same standard. But obviously,
this is not how we interpret it: we take (18b) to mean that Vernon is tall with
respect to the average adult man, while his daughter is tall with respect to the
average six-year-old.
(18)
a. Jude drank some of that, and Leo did too.
(Klein 1980)
b. Vernon is tall, and his six-year-old daughter is too.
In the end, Kennedy has to assume that pos is ambiguous: pos1 , which has the
denotation in (16b), needs a measure phrase to saturate its degree argument,
while pos2 is used with the positive form and has the denotation in (19):
(19)
a. [[pos2 ]] = λGλCλx∃d[G(d)(x) ∧ pos(d)(stnd(G)(C))]
b. [[pos(d)(stnd(G)(C))]] = 1 iff [[d]] ≥ [[stnd(G)(C)]]
Here, C is the comparison class, and stnd is a function that computes the standard value given the gradable adjective and the comparison class; for example,
given a set of individuals (the extension of C in a particular world) and the measure function tall, it returns these individuals’ mean height as the standard
degree.
In subsequent work of Kennedy, pos2 is simply called pos, while pos1 resurfaces in Svenonius and Kennedy (2006) as meas (for measure). meas imposes
14
a requirement on its complement: it has to be a function from individuals to
measurable degrees. This reflects the fact that (20a) is perfectly grammatical,
whereas (20b) is not:
(20)
a. Vernon is five feet tall.
b. *Vernon is five hours tired.
As can be seen above, pos and meas are closely related; their truth conditions
are the same (the actual degree has to be greater than or equal to the standard
degree), and the only difference is the way the standard-degree argument is
saturated.
Finally, although little hinges on this for present purposes, I will follow Kennedy
& McNally (2005) and Svenonius & Kennedy (2006) in assuming that comparatives do not themselves involve a comparison between actual and comparison
value; this function still has to be performed by the degree heads pos or meas.
Rather, comparatives alter the degree scale that comes with the adjective: the
scale of taller than Vernon, for example, is the scale of height with everything
below Vernon’s height cut off, such that Vernon’s height is the new zero degree
(more about degree scales in the next section). This division of labour fits with
Corver’s previously mentioned syntactic observation that comparative morphology (more, -er, less) is different from the ‘real’ degree morphology located in
Deg0 . A consequence of this is that the ‘degree’ morphology in Q0 cannot turn
hd, he, tii degree predicates into ordinary he, ti predicates - only Deg-heads can
do this. This observation will be relevant in Chapter 2.
To summarise these last two sections, the paradigm I am so far assuming is
the following:
(21)
a.
DegPhe,ti
Deg0 he,ti
Deg0hhd,eti,hetii
pos, how, too,
as, so
QPhd,eti
Q0hd,eti
(modifier)
Q0hhd,eti,hd,etii
(-er, more,
less, enough)
15
APhd,eti
adjective
b.
DegPhe,ti
Deg0hd,eti
QPhdi
measure phrase
Deg0hhd,eti,hd,etii
QPhd,eti
meas, too
Q0hd,eti
Q0hhd,eti,hd,etii
(-er, more, less)
APhd,eti
adjective
Note that the semantics of pos is slightly simplified here. As the way the
standard value gets resolved is beyond the concern of this thesis, I will henceforth
leave out the stnd function and the comparison class and treat the standard
value as fixed. The denotation of pos thus becomes
(22)
λGλx∃d[G(d)(x) ∧ d ≥ sG ].
1.1.4
Properties of the degree scale
So far, we have been talking about degrees as ‘points on a scale’ without really
having defined or formalised this notion. A trivial definition of a scale might
be ‘a set of degrees’. But this definition is too broad: it excludes many of our
intuitions about the nature of a ‘scale’, and neither does it provide us with a
way to account for certain properties of degree constructions.
Following Bierwisch (1989), Rotstein & Winter (2004), Kennedy (2001),
Kennedy & McNally (2005) and Rett (2008), among many others, we will define
a scale as a triple hD, R, φi, where D is a set of degrees, R a total asymmetric
ordering relation on D, and φ a dimension (i.e. what quantity is measured).
Dimension. The dimensional parameter captures the fact that not all degrees
are alike - just as one cannot measure height in kilograms, degrees on a height
scale and degrees on a mass scale are two incomparable species. This explains
the fact that (23a) is perfectly fine, while (23b) is anomalous. Both wide and
long involve degrees of linear extent, which can therefore be equated, but the
same does not hold for fat and bald.
(23)
a. A square is as wide as it is long.
b. #Maxwell is as fat as he is punctual.
16
Ordering relation. The antonyms short and tall have the same dimension
and the same set of degrees. However, they express inverse ordering relations:
the higher up a degree d is on the shortness scale, the lower it will be on the
tallness scale and vice versa. This accounts for the tautological nature of (24):
(24)
Maxwell is taller than Vernon iff Vernon is shorter than Maxwell.
Moreover, the relevance of the ordering relation parameter is demonstrated by
several contrasts between antonyms:
(25)
Compatibility with measure phrases:
a. Vernon is five feet tall.
b. #Vernon is five feet short.
(26)
Entailment of the positive form:
a. Maxwell is as tall as Vernon 6|= Maxwell and Vernon are tall
b. Maxwell is as short as Vernon |= Maxwell and Vernon are short
(27)
‘Cross-polar anomaly’:
a. Maxwell is wider than Vernon is tall.
b. #Vernon is taller than Maxwell is short.
I will not try to account for these facts here; the reader is referred to e.g.
Kennedy (2001) and references therein, Rett (2008) and Winter (2009).
Degree scales have several other properties, of which I discuss one below, as
it is related to the distribution of degree modifiers.
Open and closed scales. Cruse (1980) argues that antonymous pairs of
gradable adjectives come in two flavours, which he calls complementaries and
antonyms. A pair of adjectives is said to be complementary when its denotations
are disjoint; if A1 and A2 are complementaries, any object that is not A1 is
necessarily A2 and vice versa. In contrast, antonyms do not exhaust their
domain, such that, if B1 and B2 are antonyms, an object can be neither B1 nor
B2 . Examples of the former are clean/dirty and closed/open, while examples of
the latter are short/tall and young/old.
Furthermore, Cruse makes the important observation that antonyms are
incompatible with proportional modifiers like half and mostly, and endpointoriented modifiers like almost or fully:
(28)
a. The door is half closed/open.
b. #Vernon is half tall.
c. The table is almost/fully clean.
d. #Maxwell is almost/fully young.
17
Cruse proposes that the reason for this is that antonyms map onto unbounded
scales: as unbounded scales are infinite, it is impossible to compute some proportion of it or refer to an endpoint.8
However, complementaries do not behave uniformly with respect to the
above modifiers. Only one class of complementaries is compatible with proportional modifiers, namely, those adjective pairs where both members are
compatible with endpoint-oriented modifiers. The other class is comprised of
those adjective pairs of which only one member is compatible with an endpointoriented modifier. The first is exemplified by the pair open/closed, while the
second is exemplified by dirty/clean. Compare for example (28a) with (29a)
and (28c) with (29b):
(29)
a. #The table is half clean/dirty.
b. #The table is almost/fully dirty.
Following the same line of reasoning, this can be explained by assuming that
open/closed -type complementaries map onto scales that are bounded on both
ends, whereas complementaries like dirty/clean map onto scales that are bounded
on one end. More specifically, there is no maximum to how dirty something can
be, hence almost/fully dirty is anomalous; however, cleanliness does have a
maximum (once something has reached a certain point of cleanliness, it simply
cannot get any cleaner), which is why almost/fully clean is fine. The adjective
pairs that correspond to a scale that is bounded on one end are called partial/total pairs in the terminology of Yoon (1996) and Rotstein & Winter
(2004); for example, clean is total, as something should have no amount of dirtiness whatsoever (or: a maximal amount of cleanliness) in order to be clean,
whereas dirty is partial, as something needs to have only a certain degree of
dirtiness in order to be dirty.
Following Kennedy & McNally (2005), we will call scales that have both a
minimum and a maximum closed, and scales that have either a minimum or
a maximum, but not both, open. While this classification is related to Cruse’s
complementary/antonym distinction, it has at least three empirical benefits
over it. First, as shown above, it explains the distribution of the proportional
modifier half (which can occur with closed, but not with open scales). Secondly, it correctly predicts that certain non-complementary antonyms do allow
proportional modification:
(30)
a. The glass is not full 6|= The glass is empty
8 This
may seem strange if we consider adjectives like short and young; after all, mathematically speaking, there are clear minima to height and age (zero millimeters, zero seconds).
However, considering the fact that having a height or age of absolutely zero is actually the
same as having no height or age at all, it is clear that ‘zero’ cannot be part of the linguistic
set of degrees of height or age. But for any height or age greater than zero - even if it’s merely
a millimeter or a second - we will always be able to imagine someone who is, for example,
half that age or height. In other words, the linguistic scale of ‘shortness’ is asymptotic (so
to speak) and hence unbounded. In the words of Kennedy & McNally (2005): “Scales that
are open on the lower end include all of those degrees that approach the limit of 0 but lack a
degree whose value is less than that of all the others in the set; scales that are closed on the
lower end include such a minimal value, equal to 0”.
18
b. The glass is not empty 6|= The glass is full
c. The glass is half/almost/fully empty/full.
And finally, it accounts for the distribution of non-proportional modifiers, like
the intensifiers very and extremely, which are incompatible with closed-scale
adjectives:
(31)
a. #The door is very closed.
b. #The glass is extremely empty.
c. #The ghost is incredibly invisible.
In short, the distinction between open and closed scales is an important property of adjectival semantics, that adequately accounts for the distribution of
degree modifiers.9 10
Related to the presence of minimal and maximal degrees is the nature of the
standard value. In the paper discussed above, Cruse (1980) notes that complementaries in the positive form, unlike antonyms, do not involve comparison
with some contextually determined standard or average; rather, their meaning
can be paraphrased as ‘adjective enough’. The standard value of adjectives like
clean and safe thus corresponds to the endpoint of their scale. Adjectives with
a contextually determined standard are called relative, and adjectives with
a standard value corresponding to the minimum or maximum of their degree
scale are called absolute (Kennedy & McNally 2005).11
9 In an interesting study, Syrett & Lidz (in prep.) show that children as young as 30 months
use adverbs as a clue to adjective meaning. Children were presented with a tall, transparent
jar, and told that it was ‘very/completely/∅ pelgy’. Subsequently, they were presented with
a short, opaque jar, and told that it was ‘not pelgy’. In the actual test, they were presented
with both a short, transparent jar and a tall, opaque one, and asked ‘which one is pelgy?’ It
turned out that children in the very condition looked longer at the tall jar, while those in
the completely condition looked longer at the transparent one. The children in the no-adverb
condition did not show any preference.
10 See also Kennedy & McNally (2005), who base their account of the distribution of very,
much and well as modifiers of deverbal adjectives (e.g. needed, acquainted, surprised) on the
open/closed distinction.
11 The pair full/empty behaves differently from other closed-scale pairs (like open/closed) in
many respects. I already noted above that it is non-complementary (or: total/total rather than
total/partial), but nevertheless behaves like Cruse’s complementaries in allowing proportional
and endpoint-oriented modification. Furthermore, unlike other closed-scale adjectives, full
and empty allow modification by intensifiers like very, extremely and incredibly. However,
the entailment patterns that arise do differ:
(i)
a. Vernon is very/quite tall |= Vernon is tall
b. The glass is very full 6|= The glass is full
c. The theater is quite empty 6|= The theater is empty (cf. Rotstein & Winter 2004)
My tentative suggestion is that the existence of a ‘grey area’ in the middle of the full/empty
scale (due to the total nature of both full and empty, the largest part of the scale actually
corresponds to objects that are neither) somehow leaves a pragmatic space for a relative
interpretation, which can be triggered by modification or certain contextual factors - a scale
measuring ‘volume of content’ rather than ‘fullness/emptiness’. This means that full and
empty can behave in both an absolute and a relative way. How this suggestion holds up under
further theoretical and empirical scrutiny remains to be seen.
19
1.2
The semantics of degree modif iers: some issues
If we assume that DegP looks like (21) above - and we have good reason to - we
have to treat the modifier in SpecQP as a function hhd, eti, hd, etii from degree
predicates to degree predicates.12 In other words, if we modify tall by extremely,
we get a degree predicate that relates an individual and a degree of ‘extreme
tallness’. Subsequently, pos gets applied to it in the usual way, which explains
why even modifiers that indicate a small degree, like slightly, result in a positive
reading. This is similar to the treatment of comparatives I briefly discussed a
few pages back: the result is, essentially, a new, absolute scale with the smallest
degree that can still be called ‘extreme’ as its minimum. The incompatibility
of this new measure function with a measure phrase - as exemplified by the
ungrammaticality of a sentence like (32) - would be explained by the fact that
degrees of ‘extreme tallness’ are not measurable (in the sense of Svenonius &
Kennedy 2006) the way ordinary degrees of tallness are.
(32)
*Maxwell is seven feet extremely tall.
However, while this seems an unproblematic assumption for modifiers like very
or Dutch ontzettend, which (almost) exclusively occur as degree modifiers (ontzettend originally means horrifying, but is hardly used as such anymore), it is
much less elegant in the case of evaluative adverbs. As Morzycki (2004) notes,
the evaluative adverbs that can function as degree modifiers have the property
that they can also occur in clause-modifying positions (examples from Morzycki
2004):
(33)
a. Clyde is remarkably tall.
b. Remarkably, Clyde is tall.
(34)
a. Floyd is surprisingly ugly.
b. Surprisingly, Floyd is ugly.
(35)
a. Floyd’s SUV is unacceptably inefficient.
b. Unacceptably, Floyd’s SUV is inefficient.
12 A
second possibility is put forward by Klein (1980), who analyses degree modifiers as
in fact modifying comparison classes. Thus, roughly, very in very tall picks out from the
original comparison class only those individuals that are tall and returns this set as the new
comparison class; hence the meaning of Vernon is very tall can be paraphrased as ‘Vernon
is tall for a tall guy’. I will not pursue this option further here, because it seems to me very
hard to ground in the syntax of DegP as presented in this chapter - Klein’s analysis is purely
semantic.
20
To account for this fact, we would have to assume that all these adverbs13 are
systematically ambiguous between a measure phrase modifier and a propositional modifier; as these are rather different semantic operations, it would seem
extremely hard to come up with a semantics for these adverbs that is capable
of doing both. But of course, we don’t want to assume of an entire open class
of words that they are all ambiguous.
Another problem for the above account is our earlier observation (in section 1.1.2) that many of the modifiers can themselves be modified, even by
each other. Examples like surprisingly shockingly rude or unusually endearingly
helpful may not be perfectly ordinary, but they are perfectly interpretable. If
these adverbs were of type hhd, eti, hd, etii, we would not expect them to be able
to modify each other (also in section 1.1.2, we ruled out the option of intersective
modification).
Finally, an analysis of modifiers as scale-altering functions from degree predicates to degree predicates does no justice to the wide range of existing degree
modifiers. It might seem to work fine for modifiers like very, pretty and fairly,
but if all surprisingly, incredibly and blood-curdlingly were able to do was redefine the minimum of the scale (or, in Klein’s analysis, pick out some subset of the comparison class), it would be a mystery how to account for their
(differences in) intuitive meaning without ending up with some kind of circularity (e.g. analysing Vernon is surprisingly tall as ‘Vernon is tall for a surprisingly
tall guy’).
The above seems to call for a non-uniform treatment of degree modifiers. On
the one hand, there are modifiers like very, pretty, fairly and Dutch ontzettend,
which can only modify degree and are non-gradable. The analysis outlined
above seems to work fine for them.14 On the other hand, there are modifiers
like our evaluative adverbs, that can modify elements other than degree, are
gradable, and form a problem for the analysis above.
One approach in which the two classes of modifiers receive a non-uniform
treatment is Morzycki (2004), who proposes that the latter class of adverbs,
like measure phrases, appear in SpecDegP where they are licensed by a covert
degree head r. Semantically, they denote simple propositional modifiers; the
degree-modifying interpretation results from the semantics of r. The advantage of his analysis is that the adverbs have precisely the same denotation as
13 Morzycki’s
(2004) paper includes the following small sample: amazingly, astoundingly,
arousingly, calmingly, disappointingly, earth-shatteringly, excitingly, extraordinarily, frighteningly, grotesquely, heart-breakingly, horribly, illegally, impressively, inappropriately, inconceivably, infuriatingly, interestingly, irritatingly, laughably, mind-numbingly, nauseatingly,
provocatively, revoltingly, ridiculously, satisfyingly, shockingly, stimulatingly, stunningly,
suf f iciently, terribly, terrifyingly, typically, (un)acceptably, unbelievably, unexpectedly, unnervingly, (un)pleasantly, (un)remarkably, unusually, upsettingly, uselessly, and wonderfully.
14 Ontzettend is interesting from this point of view. Although its original evaluative meaning
of ‘shocking, horrifying’ is negative, it is purely neutral in its use as a degree modifier (one
could make the same claim about English terrible, although not all native speakers I consulted
agree that this can be used neutrally). This fits with an analysis in which ‘pure’ degree
modifiers only function to alter the adjective’s degree scale, which is essentially a neutral
operation - there is no way to contribute anything else to the semantics.
21
their corresponding adjective; this makes it a good starting point for a parallel
theory on noun gradability. There are also some problems with it, however.
For example, the SpecDegP location leads to several wrong predictions. Also,
the syntactic and semantic similarities that do exist between the two classes of
adverbs are ignored and obscured by the completely different treatment they
receive in Morzycki’s system. In the following chapter, I will look at Morzycki’s
analysis in more detail.
22
Chapter 2
Evaluative adverbs as
degree modifiers
1
In this chapter, I will present an analysis of degree modification by evaluative
adverbs (which I will henceforth refer to as EDC, for ‘evaluative degree construction’). Section 2.1 summarises one of the few, if not the only, existing accounts of
EDCs, as developed in Morzycki (2004)2 . In section 2.2, I will argue in favour of
a simplification of the semantics proposed by Morzycki. Subsequently, in section
2.3, I will demonstrate that Morzycki’s analysis is syntactically inaccurate, and
present an alternative analysis based on data that show so-called much-support
(Corver 1997a), and on gender agreement phenomena in Dutch. Finally, I will
show how the new syntax and semantics fit together.
2.1
Previous accounts: Morzycki (2004)
Morzycki’s starting point is the observation that a large and open class of adverbs, which he calls remarkably-adverbs, can systematically modify gradable
adjectives as well as complete sentences. The different positions are associated
with a clear difference in meaning:



 surprisingly 



remarkably
(1)
a. Maxwell is
tall.
amazingly 





shockingly
1 A paper based on this chapter will be presented at the Student Session of the 2010
European Summer School in Logic, Language and Information (ESSLLI) in Copenhagen. I
am grateful to the two anonymous reviewers for valuable comments and suggestions, which
the paper as well as this chapter has definitely benefited from.
2 Another one is Katz (2005), although he only discusses the semantic, not the syntactic,
side of the matter
23


Surprisingly 





Remarkably
b.
, Maxwell is tall.

 Amazingly 



Shockingly
The sentences in (1a) do not entail those in (1b): if we were expecting Maxwell
to be tall, but just not that tall, we could utter (1a) but not (1b).
The semantics of the (b)-sentences is uncomplicated: the adverb simply
modifies the proposition expressed by Maxwell is tall. But what exactly do the
adverbs modify in an EDC like Maxwell is remarkably tall ? Do they similarly
modify propositions, and if so, what do these propositions express? Where does
the semantic difference between (1a) and (1b) come from? Morzycki considers
several paraphrases:
(2)
Maxwell is remarkably tall.
a. It is remarkable that Maxwell is as tall as he is.
b. It is remarkable to be as tall as Maxwell is.
c. It is remarkable how tall Maxwell is.
He subsequently rejects (2a) and (2b) for two reasons: they are true in several situations in which we would understand the original sentence to be false,
and unlike the original sentence, they are not factive - they do not entail that
Maxwell is, in fact, tall. To make the first point clear, consider a situation in
which Maxwell is in fact remarkably short - surely, we would be able to claim
(2a). However, we would not call the remarkably short Maxwell remarkably
tall. In another scenario envisioned by Morzycki, Maxwell was born at precisely
5:09 in the morning, on the fifth day of the ninth month of 1959 - and to our
amazement, his height happens to be exactly five feet and nine inches. This is
remarkable indeed, and yet, again, we would not be able to claim that Maxwell
is remarkably tall.
The second point is related to the first: for the paraphrases in (2a) and
(2b) to be true, there is no need for Maxwell to be in fact tall. However, as
Morzycki argues, the sentences in (1a) all entail their corresponding positive
form: if Maxwell is remarkably tall, it must also be true that he is tall.
The remaining paraphrase - that Morzycki takes to be the correct one - is
(2c), in which the adverb modifies an embedded exclamative (How tall Maxwell
is! ).
Following Zanuttini & Portner (2003), Morzycki takes exclamatives to denote
sets of true propositions, just as questions do (cf. Karttunen 1977):
(3)
[[How tall Maxwell is! ]] = {p : p is true and there is a degree of height d
such that p is the proposition that Maxwell is d-tall}
The type of an evaluative adverb like remarkable is a function hhhs, ti, ti, ti from
sets of propositions hhs, ti, ti to truth values. For Maxwell is remarkably tall to
be true, remarkable(p) must be true for some proposition p in the set (see
(4b)). For current purposes, this amounts to the denotation in (4c), in which
reference is made to sets of degrees rather than sets of propositions.
24
(4)
a. Maxwell is remarkably tall = It is remarkable [how tall Maxwell is!]
b. ∃p[p ∈ {‘Maxwell is 6 feet 1 inch tall’, ..., ‘Maxwell is 6 feet 2 inches
tall’, ..., ‘Maxwell is 6 feet 3 inches tall’, ..., ‘Maxwell is 6 feet 4 inches
tall’} ∧ remarkable(p)]
c. remarkable(ˆ∃d[d ∈ {‘6 feet 1 inch’, ..., ‘6 feet 2 inches’, ..., ‘6 feet
3 inches’, ..., ‘6 feet 4 inches’} ∧ Maxwell is d-tall])
In short, analysing constructions like the one in (4a) involves quantification
over a set of degrees. But how is this set constructed? To answer this question, Morzycki again looks at exclamatives. Two important properties of the
semantics of exclamatives, according to Zanuttini & Portner, are factivity
(see above) and domain widening. To understand the nature of the latter,
consider the difference between (5a) and (5b) (from Morzycki 2004):
(5)
a. Herman eats everything.
b. What (surprising) things Herman eats!
Arguably, the domain of everything in (5a) is restricted by the context such
that we do not expect it to include “lightbulbs, his relatives, or presidential
elections” - or, in general, anything but ordinary food. For (5a) to be true, it
is not necessary that Herman’s eating habits include things like live locusts for
breakfast; it merely suggests that Herman is a particularly easy dinner guest.
In contrast, (5b) does suggest that the domain of things eaten by Herman also
includes the extraordinary, like live locusts or raw serrano chillies. This is the
effect of domain widening.
To adequately capture the semantics of EDCs, then, it is necessary to incorporate both domain widening and factivity into the semantics of the embedded
exclamative. This ultimately leads Morzycki to propose the following semantics
for Maxwell is remarkably tall :
(6)
remarkable(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C∧ tall(d)(m) ∧d ≥ stall ])
Here, factivity is captured by d ≥ stall , and domain widening by C 0 ⊃ C ∧ d ∈
C 0 − C. The former in fact amounts to incorporating the semantics of pos (as
repeated in (7); see also section 1.1.3) into the denotation:
(7)
[[pos]] = λGλx[∃d[G(d)(x) ∧ d ≥ sG ]]
The latter probably needs more explanation. It ensures that there is some
widened domain C 0 that is a superset of the original domain C, and moreover,
that the degree d to which Maxwell is tall is contained in that part of the
domain which was created by the widening (i.e. C 0 − C). After all, only this
part includes the degrees that do in fact correspond to a remarkable height.
We might wonder why it is still necessary to build in a contextual restriction
on d by means of C and C 0 , if the pos-part already ensures that Maxwell’s
degree of height exceeds the standard. But leaving this part out would lead to
the wrong semantics:
25
(8)
remarkable(ˆ∃d[ tall(d)(m) ∧d ≥ stall ])
(8) amounts to stating that it is remarkable that Maxwell is tall, which is obviously not what we are trying to say. In contrast, the denotation in (6) not only
captures the fact that Maxwell’s degree of height must exceed the standard,
but also that it must fall within the domain widened by exclamativity; in other
words, that it must be a degree that is somehow unexpected, “not among [those]
already contextually salient” (Morzycki 2004).
Morzycki’s next goal is to provide a syntax for EDCs that will lead to the semantics in (6). He proposes that evaluative modifiers (i.e. remarkably-adverbs)
are located in SpecDegP, where they are licensed by a covert degree head. This
means that evaluative modifiers are treated similar to measure phrases, which
also occur in SpecDegP and are licensed by a covert degree morpheme in D0 .
This, according to Morzycki, would account for certain distributional similarities
between measure phrases and modifiers (examples from Morzycki 2004):
six feet
*six feet
(9)
a. Floyd is
tall
.
remarkably
*remarkably
*six feet
b. Floyd is
very tall.
*remarkably
Morzycki chooses to treat the evaluative modifiers themselves as simple propositional modifiers, as in (10), and have the semantics of the degree head - which
he calls r - take responsibility for factivity and domain widening ((11b)).
(10)
[[remarkably]] = [[remarkable]] = λp[remarkable(p)]
(11)
[[r]] = λGλRλx[R(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ G(d)(x) ∧ d ≥ sG ])]
In building up the semantics of an EDC like Maxwell is remarkably tall, the head
r is applied to the denotations of its complement (AP) and specifier (remarkably, or in general, AdvP), and the resulting IP-denotation is in turn applied
to the denotation of its specifier (Maxwell, the subject DP). The following tree
and step-by-step derivation show how this process results in the semantics established in (6):
(12)
IP
VP
DP
Maxwell
V
is
DegP
Deg0
AdvP
remarkably
D0
AP
r
tall
26
(13)
[[Maxwell is remarkably r tall ]]
= [[r]]([[tall ]])([[remarkably]]))([[Maxwell ]])
= λGλRλx[R(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ G(d)(x) ∧ d ≥ sG ])]
(λdλx[tall(d)(x)])([[remarkably]]))([[Maxwell ]])
= λRλx[R(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ tall(d)(x) ∧ d ≥ stall ])]
(λp[remarkable(p)])([[Maxwell ]])
= λx[remarkable(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ tall(d)(x)∧
d ≥ stall ])](m)
= remarkable(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ tall(d)(m) ∧ d ≥ stall ])
The remainder of this chapter will be devoted to a critical evaluation of
Morzycki’s analysis. Based on these criticisms, I will propose an alternative
analysis that I feel is an improvement in several ways. However, it is still similar
in spirit to Morzycki’s analysis as summarised above, in retaining the notion of
a covert degree head r and remarkably-adverbs as propositional modifiers.
2.2
Semantics
In this section, I will argue that analysing EDCs as embedded exclamatives leads
to unnecessary complications. I also show that Morzycki’s reasons for rejecting
the alternative cease to be valid when we take into account the fact that both
gradable predicates and evaluative adverbs are monotone. Because of this, I
argue that the semantics of r can be much simpler.
2.2.1
Some arguments against EDCs as embedded exclamatives
Morzycki’s EDC denotation - repeated for ease of reference in (14) - has the
evaluative modifier taking scope over the existential quantifier:
(14)
[[EDC]] = R(ˆ∃d∃C 0 [C 0 ⊃ C ∧ d ∈ C 0 − C ∧ G(d)(x) ∧ d ≥ sG ])] (for some
gradable adjective G, evaluative adverb R, domain C and individual x)
This is a consequence of treating EDCs as exclamatives embedded under a
modifier: the existential quantifier is part of the exclamative and hence has
to be interpreted inside the scope of the modifier. There is something not
entirely satisfying about this. If quantification over d would take place outside
of the scope of the modifier, we would obtain a denotation that seems to fit the
semantics of EDCs remarkably well, without having to involve C and C 0 :
(15)
[[EDC]] = λxλG∃d[R(ˆG(d)(x)) ∧ G(d)(x) ∧ d ≥ sG ]
Indeed, there is something redundant about the domain widening part of the
denotation. It guarantees that the degree to which x is A is somehow so ‘extreme’ that it falls outside of the range of degrees we would naturally consider.
27
But that is just another way of saying that what is going on is ‘remarkable’, or
‘surprising’, or ‘unbelievable’. As an illustration, take a sentence like Maxwell
is remarkably tall. Paraphrasing Morzycki’s denotation, the semantics of this
would boil down to something like ‘It is remarkable that Maxwell’s degree of
tallness is such that it is somehow unexpected’.
This is, in fact, a general problem of analysing EDCs as embedded exclamatives. An exclamative (How tall Maxwell is! ) can itself be paraphrased
as something roughly like ‘Maxwell is unexpectedly tall’. To quote Morzycki
himself:
[The] idea [of domain widening] elegantly gathers together several
otherwise slippery and elusive intuitions about what exclamatives
mean. Among these are the intuition that exclamatives somehow
involve an ‘extreme’ value for something, and that exclamatives convey that something is unexpected in a particular way.
Embedding this under a modifier that also conveys a sense of unexpectedness
or surprise, then, leads to a strange kind of redundancy. After all, it suggests
that the ‘unexpectedness’ of Maxwell being d-tall is the case even before the
contribution of remarkably to the semantics. All in all, resorting to C and C 0
seems merely a clever way to smuggle the semantics of remarkable into the scope
of the quantifier, rather than something independently motivated.
Considering the above, we can test empirically whether the embedded-exclamative analysis is true: we would expect the sense of unexpectedness or extremeness caused by domain widening to be there, regardless of the meaning of the
modifier. This expectation is not borne out, however. The evaluative adverbs
in (16) themselves do not express anything ‘extreme’, and indeed, the sentences
in (16) do not seem to suggest unexpectedness or extremeness in any way.


 disappointingly 
arousingly
tall.
(16) Vernon is


satisfyingly
In short: the domain widening effect of certain EDCs, like Maxwell is remarkably tall, seems to be a consequence of the semantics of the particular adverb,
rather than a property of this kind of construction in general. In light of this,
it is unjustifiably redundant to incorporate domain widening into the EDC denotation separately from the denotation of the adverb.
Considering all this, it is clear that a semantics without domain widening is
to be preferred. As we have seen in the previous section, we cannot just scratch
the domain widening part from the denotation, as this leads to the wrong interpretation. What we need, therefore, is a semantics in which the existential
quantifier binding the degree variable has scope over the evaluative adverb (as
in (15)). As can be seen in (17), this defines the relationship between the different semantics associated with different adverb positions in an elegant, intuitive
way, namely in terms of scope:
28
(17)
a. Maxwell is remarkably tall:
∃d[remarkable(ˆtall(d)(m)) ∧ . . .) . . .]
b. Remarkably, Maxwell is tall:
remarkable(ˆ∃d[tall(d)(m) ∧ . . .]) ∧ . . .
However, one problem remains: the semantics we have thus arrived at corresponds to one of the paraphrases that Morzycki (2004) found reason to reject.
The following section explores whether he was right to do so.
2.2.2
In which monotonicity saves the day
Consider again the data in (18):
(18)
a. ‘Clyde is remarkably tall.’
b. The degree to which Clyde is tall is remarkable (cf. Cresswell 1976,
Katz 2005)
c. Clyde is in fact remarkably short.
d. Clyde’s height is exactly five feet and nine inches; moreover, he was
born at precisely 5:09 in the morning, on the fifth day of the ninth
month of 1959.
Morzycki (2004) claims that (18b) (or, more formally, ‘There is a degree d such
that Clyde is d-tall and it is remarkable that Clyde is d-tall’) cannot be an
accurate representation of the semantics of (18a), as it would be true in cases
(18c-d), which obviously are not captured under the actual meaning of (18a).
Recall that this is Morzycki’s main reason for deciding to treat sentences like
18a) as if they contained an embedded exclamative (‘It is remarkable [how tall
Clyde is!]’).
However, as Katz (2005) and Nouwen (2005) show, this objection to the
paraphrase in (18b) does not hold. Following them and Heim (2000), we assume
that gradable predicates are monotone in the following sense (Heim 2000):
(19)
A function f of type hd, he, tii is monotone iff
∀x∀d∀d0 [f (d)(x) = 1 & d0 < d → f (d0 )(x) = 1]
In words: If x has a certain property to a degree d, it also has this property to
all lower degrees d0 .
If we assume that tall is monotone, (19a) - along with the paraphrase in
(19b) - is correctly ruled out as an appropriate description of the situations in
(19c-d), as it entails that any degree of height d0 > d would also have been
remarkable (Nouwen 2005, Katz 2005). For (19c), that means that Clyde’s
height would still have been remarkable had it been average (i.e. taller than
remarkably short), and for (19d), that a height of six feet would have been as
remarkable as Clyde’s actual height of five feet nine. Obviously, neither is the
case.
To see why this entailment holds, consider the fact that the modifier remarkably is also monotone - downward monotone, to be specific. A downward
29
monotone operator O, when applied to some proposition p, reverses p’s entailments:
(20)
a. O is MON↓ iff p |= p0 ⇒ O(p0 ) |= O(p)
b. Mildred is reading a Booker Prize-winning novel |= Mildred is reading
a novel
c. It is remarkable for Mildred to be reading a novel |= It is remarkable
for Mildred to be reading a Booker Prize-winning novel
Similarly, as the monotonicity of tall implies that tall(d)(x) |= tall(d0 )(x)
where d0 ≤ d,
(21)
remarkable(ˆtall(d0 )(x)) |=remarkable(ˆtall(d ≥ d0 )(x))
In other words, the monotonicity of both remarkable/remarkably and tall
ensures that if it is remarkable that x is d0 -tall, x being d ≥ d0 -tall must also be
remarkable. Hence, there is an independent reason why freakish heights (like one
coinciding with one’s date and time of birth) can not be called ‘remarkable’.3
This is the case regardless of the paraphrase we choose, so contrary to what
Morzycki claims, we are not obliged to pursue an embedded-exclamative analysis
because of this.
Factivity, however, is another matter. If we take Maxwell is remarkably tall
to have a semantics along the lines of ‘There is a degree d such that Maxwell
is d-tall and it is remarkable that Maxwell is d-tall’, the positive form Maxwell
is tall is not entailed. However, we can easily remedy this by extending our
paraphrase a bit: ‘There is a degree d such that d exceeds the relevant standard
value and Maxwell is d-tall and it is remarkable that Maxwell is d-tall’.4
Concludingly, we can use the EDC denotation arrived at in section 2.2.1
without landing ourselves into trouble. Considering the way the semantics of
3 Nouwen (2009) uses monotonicity to account for another puzzling observation, originally
made by Zwicky (1970), about the behaviour of adverb pairs like usually/unusually in sentences like The children are (un)usually noisy. More about this in section 3.1.
4 It can also be argued that evaluative degree constructions do not actually entail their
corresponding positive form. For example, it is not obvious that (ia) should entail (ib):
(i)
a. Vernon’s bathroom is surprisingly clean.
b. Vernon’s bathroom is clean.
However, the judgements are complicated here, as even total adjectives like clean are not
always completely black and white. It might well be that (ia) does not entail (ib) because their
comparison classes are different, as intuitively seems to be the case: in (ia), the cleanliness of
Vernon’s bathroom seems to be evaluated with respect to what we expect of Vernon, perhaps
based on the state of other rooms in his house, while in (ib), Vernon’s bathroom seems to be
judged against some general standard of bathroom cleanliness. If we make the comparison
class explicit, (ia) does seem to entail the corresponding positive form:
(ii)
a. Vernon’s bathroom is suprisingly clean (for a room inhabited or used by Vernon).
b. Vernon’s bathroom is clean (for a room inhabited or used by Vernon).
However, if for some reason one would still want to reject the entailment of the positive form
by evaluative degree constructions, it is pretty easy to do so based on the present analysis:
just scratch the relevant part of the EDC semantics.
30
EDCs is derived in Morzycki’s analysis, the obvious way to carry out this change
is to adjust the semantics of r. After all, the other ingredients of the denotation
do not change, they are just assembled in a somewhat different way.
The denotation I propose for r is the following:
(22)
[[r]] = λGλRλx[∃d[G(d)(x) ∧ d ≥ sG ∧ R(ˆG(d)(x))]]
Here, G is a gradable adjective, R an evaluative adverb, and sG the standard
value for G with respect to a contextually defined comparison class.
2.3
Syntax
In this section, I will argue that evaluative modifiers cannot be treated on a par
with measure phrases, as Morzycki does in assigning them both to SpecDegP.
Rather, evaluative modifiers are - just as other degree modifiers, like very and
pretty - located in SpecQP, and r in Q0 . Evidence based on much-support and
gender agreement in Dutch is presented to support this claim.
2.3.1
Not in SpecDegP, but in SpecQP
First of all, it must be noted that modifiers do not pattern with measure phrases
in many respects. For example, Morzycki claims on the basis of (23) that both
measure phrases and modifiers are “in complementary distribution with overt
degree words in the [DegP] in which they occur”, which he takes as evidence
that both must be licensed by a covert degree head.
*six feet
(23) Floyd is
very tall.
*remarkably
But very, as we have seen in section 1.1.2, is not a degree head but a modifier. Six feet and remarkably are ruled out for two completely different reasons
here: six feet because measure phrases and modifiers are generally incompatible
(cf. section 2.4.1), and remarkably because the only available modifier position, SpecQP, is taken by very (and very, being non-gradable, cannot itself be
modified).
In fact, as we have already seen in section 1.1.2, evaluative modifiers (like
other degree modifiers, and very much unlike measure phrases) are in fact perfectly grammatical if they appear to the right rather than to the left of the
degree head.
*six feet
(24)
a. How
tall Maxwell is!
remarkably
*six feet
b. Maxwell is so
tall that all tourists want to take a
remarkably
picture with him.
*six feet
c. Maxwell is just as
tall as Vernon.
remarkably
31
One might object that it is not clear whether the various degree heads here in
fact head the DegP that contains remarkably as a modifier. They might also
modify remarkably itself, such that [how/so/as remarkably] as a whole is the
modifier, and DegP is still headed by pos. However, there are at least three
reasons to assume that this is not the case. First, note that we can replace each
instance of remarkably in (24) with the non-gradable very. This shows that
there is nothing wrong with having both an overt degree head and a modifier,
so as long as one cannot prove that evaluative modifiers behave differently from
very, one has no reason to assume that they do.
Secondly, recall that comparative and equative constructions are semantically anomalous if the adjectives that are compared are not measured along the
same dimension ((25a) vs. (25b)). We can use this fact to determine which
element is compared to which in a sentence like (24c). If the evaluative adverb
were the modifier in a DegP headed by as, rather than its complement, we would
expect the (a)nomaly of the comparison to depend only on the dimension of the
adjective. And in fact, this expectation is borne out:
(25)
a. Maxwell is just as tall as Vernon is wide.
b. #Maxwell is as tall as he is arrogant.
(26)
a. Maxwell is as remarkably tall as Vernon is remarkably wide.
b. #Maxwell is as remarkably tall as Vernon is remarkably arrogant.
The difference between (26a) and (26b) would be inexplicable if the first instance of remarkably were compared with the second one, or with wide/arrogant;
the only way to explain why (26b) is anomalous while (26a) is not, is to say
that (26b) compares tall and arrogant, which do not have identical dimensions,
whereas (26a) compares wide and tall, which do. This means that tall is the
complement of as, and not remarkably.
Finally, the assumption that evaluative modifiers occupy the same position
as ordinary degree modifiers like very is supported by the fact that there seems
to exist a certain diachronic continuum between them, with evaluative modifiers
losing their original meaning and coming to function as ‘pure’ degree modifiers.
Perhaps the best example is the earlier mentioned Dutch ontzettend ; English
terrible seems to be undergoing the same development. This process is much
easier to understand if the two classes share their syntactic distribution.
Concludingly, while there is no decisive way to prove for all possible overt
degree heads that they can co-occur with an evaluative modifier in SpecQP,
the facts we do have (as, at least, can; plus, evaluative modifiers pattern with
very in every other respect) are suggestive enough. I will therefore assume that
evaluative modifiers occupy the same location as degree modifiers like very and
pretty: the Spec of QP. If this is true, we expect r - which licenses evaluative
modifiers - to occur in Q0 rather than Deg0 . In the following section, I will
present some data that support this expectation.
32
2.3.2
r as a covert Q-head: evidence from much-support
For easy reference, (26) reproduces the complete extended AP as discussed in
the introduction:
(27)
DegP
Deg0
Spec
Deg0
{how, too, as, so,
pos, meas}
QP
Q0
Spec
Q0
AP
{-er, more, less, enough}
A0
A0
Furthermore, I follow Corver (1997a,b) in assuming A0 -to-Q0 movement, which
occurs when Q0 is empty or contains -er.
However, A0 -to-Q0 movement is not always possible. In the case of English so-pronominalisation, for example, where the whole AP is replaced by the
pro-form so, there is no A0 to raise to Q0 . Corver notes that this results in
ungrammaticality when Q is empty (examples from Corver 1997a):
(28)
a. John is fond of Mary. Bill seems [less so].
b. Of all careless people, no one is [more so than Bill].
(29)
a. John is fond of Mary. *Maybe he is [too so].
b. *John told me he was afraid of spiders, but I wonder [how so] he
really is.
While (28) shows that so is fine when it occurs with a Q-head (like less and
more), (29) makes clear that Q0 cannot be empty - so cannot occur when there
is only a Deg-head (like too and how ). To make the Deg0 + so combination
grammatical, Q0 needs to be nonempty, which is achieved by inserting much:
(30)
a. John is fond of Mary. Maybe he is [too much so].
b. John told me he was afraid of spiders, but I wonder [how much so]
he really is.
33
Note that the function of much here5 is purely syntactic: it does not contribute
anything to the semantics of the sentence and can therefore be rightfully called
a ‘dummy’.
Now, consider the data in (31)


remarkably 






 surprisingly 

incredibly
(31) Vernon is tall, even
(*much) so.




eerily






...
The fact that much-insertion is ungrammatical here can only be explained by
assuming that Q0 is not empty - it must be occupied by a covert element. The
obvious candidate for this is r.
There is another argument in support of this. If r’s position is indeed Q0 , we
would expect a contrast with a ‘natural-born’ degree modifier like very, which
presumeably does not need the mediation of something like r to be able to
function as such. If very can modify degree directly, Q0 will be empty, and we
would expect much-support to be necessary in the case of so-pronominalisation.
This expectation is indeed borne out (and similarly for pretty)6
very
(32) Vernon is tall, even
*(much) so.
pretty
Concludingly, the presence of a covert Q-head can be detected by testing whether
much-support is necessary or not. In the case of evaluative modifiers, such a
covert head is present; in the case of true degree modifiers like very and pretty,
it isn’t. As the difference lies in the fact that the former need mediation in order
to be able to modify degree while the latter do not, the obvious conclusion is
that this mediation is exactly the function of the covert Q-head - in other words,
r is located in Q0 .
2.3.3
r as a covert Q-head: evidence from Dutch gender
agreement
The analysis presented above defines the difference between evaluative modifiers
and ‘true’ degree modifiers (e.g. very, pretty) in terms of their ability to modify
degree directly - the latter can, while the former need the mediation of eval.
In English, this claim is supported by the presence or absence of much-support
5 Much
(iii)
can also be a lexical quantifier, as in the following examples:
a. Vernon and Maxwell resemble each other much.
b. Mildred is much afraid of guinea pigs.
c. Too much love will kill you.
This is not the use of much we are concerned with here, however.
6 So, Morzycki is correct in assuming that very and pretty are different from evaluative
modifiers, and in recognising that they form a rather closed class - this just does not mean
that therefore, they must be degree heads.
34
in the case of so-pronominalisation (very and pretty do need much-support).
Dutch does not have so-pronominalisation, but it does offer some interesting
independent evidence in the form of gender agreement between adjective and
modifier.
(33)
a. Een belachelijk(*e) dure fiets
‘a ridiculously(-INFL) expensive bike’
b. Een ontzettend(?e) mooie fiets
‘an extremely(-INFL) beautiful bike’
c. Een ?heel/hele mooie fiets
‘a very/very-INFL beautiful bike’
Belachelijke ‘ridiculous’ in (33a) cannot receive a degree-modifying interpretation; only the non-inflected form can. The use of the inflected form onzettende
‘extremely’ as a degree modifier, however, is relatively common; finally, degreemodifying hele ‘very’ has an overwhelming tendency to agree with the adjective.
Other evaluatives pattern with belachelijk. The table below shows the Google
results for several combinations of modifier and adjective:
belachelijk
belachelijke
ontzettend
ontzettende
heel
hele
dure
13,900
1,610
5,520
351
8,900
27,200
mooie
3,210
262
23,100
27,300
257,000
873,000
Assuming that agreement reflects a Spec-Head relationship, the difference follows naturally from our assumptions: EDCs do not involve a Spec-Head relationship between the modifier and the adjective, as the presence of eval in
Q0 prevents the adjective from raising there. In contrast, heel/hele ‘very’, as
a proper degree modifier, does not need the mediation of an element like eval
in Q0 , so the adjective can raise to this position, ending up in a Spec-Head
relationship with the degree modifier (cf. Corver 1997b). Finally, the mixed
behaviour of ontzettend is exactly what we would expect of an evaluative that
is diachronically turning into a ‘real’ degree modifier.
2.4
2.4.1
Assembling the pieces
Splitting up r
One final syntactic question still needs answering. If r is indeed located in Q0 , is
anything else heading DegP? This question, I think, cannot be answered without
considering the semantics of r as proposed in section 2.2, which is repeated in
(34) below.
35
(34)
[[r]] = λGλRλx[∃d[G(d)(x) ∧ d ≥ sG ∧ R(ˆG(d)(x))]]
Like Morzycki’s r, this denotation incorporates the denotation of pos, which
occurs as a degree head in the positive form (Vernon is pos tall ) and in comparatives (Vernon is pos taller than Maxwell )7 . However, as we saw in section
2.3.1, EDCs can be headed by overt degree heads in Deg0 . Recall that all degree
heads, whatever their particular semantics may be, share a single purpose: to
quantify over the adjective’s degree argument and thereby turn a hd, eti degree
predicate into an ordinary he, ti one. But in its current form, this task is already performed by r. This means that there is no way any degree head could
combine with a QP headed by r.
Furthermore, in section 1.1.3, I have briefly argued that “the ‘degree’ morphology in Q0 cannot turn hd, he, tii degree predicates into ordinary he, ti predicates - only Deg-heads can do this”. This is also incompatible with an r-head
that incorporates pos.
Considering all this, it seems clear that we should separate pos from the rest
of r and assign it to Deg0 , parallel with the aforementioned degree constructions.
Separated, the denotations of pos and r are as follows (to avoid confusion, I
will henceforth refer to r-without-pos as eval):
a. [[eval]] = λGλRλdλx[G(d)(x) ∧ R(ˆG(d)(x))]
(35)
b. [[pos]] = λGλx[∃d[G(d)(x) ∧ d ≥ sG ]]
In section 1.2, we wondered how to analyse both evaluative modifiers and ‘true’
degree modifiers - like very - in a way that would explain both their differences
and their similarities. The analysis we have currently arrived at seems very
promising: it keeps the syntactic parallel between both classes of modifier intact,
while allowing them to have quite a different semantics.
(36)
a.
DegP
Deg0
Deg
QPhd,eti
pos
Q0
evaluatively hst,ti
Q
APhd,eti
eval
7 Though this depends on the particular analysis of comparatives one wants to adopt.
Comparatives with pos are proposed in Kennedy & McNally (2005), Svenonius & Kennedy
(2006); other analyses (e.g. Heim 2000, Kennedy 2004) do not make use of pos.
36
b.
DegP
Deg0
Deg
pos
QPhd,eti
Q0
very hhd,eti,hd,eti
Q
APhd,eti
Ai
ti
Under the present analysis, both constructions are headed by pos; the only
thing that makes them different is the presence or absence of eval (and its
direct consequences, i.e. whether the adjective moves to Q0 ).
Assuming separate eval and pos heads, the semantics of a sentence like Vernon
is surprisingly tall is built up as follows:
(37)
IP
DP
Vernon
!
VP
V
DegP
is
Deg0
Deg
QP
λGλx[∃d[G(d)(x) ∧ d ≥ sG ]]
Q0
AdvP
surprisingly
Q
λGλRλdλx[G(d)(x) ∧ R(ˆG(d)(x))]
(38)
AP
λdλx[tall(d)(x)]
[[Vernon is pos eval surprisingly tall ]]
= [[pos]]([[eval]]([[tall ]])([[surprisingly]]))([[Vernon]])
= [[pos]](λGλRλdλx[G(d)(x)∧R(ˆG(d)(x))](λdλx[tall(d)(x)])([[surprisingly]]))
([[Vernon]])
= [[pos]](λRλdλx[tall(d)(x)∧R(ˆtall(d)(x)](λp[surprising(p)])([[Vernon]])
= λGλx[∃d[G(d)(x)∧d ≥ sG ]](λdλx[tall(d)(x)∧surprising(ˆtall(d)(x))])
([[Vernon]])
37
= λx[∃d[tall(d)(x) ∧ surprising(ˆtall(d)(x)) ∧ d ≥ st all]](v)
= ∃d[tall(d)(x) ∧ surprising(ˆtall(d)(v)) ∧ d ≥ st all]]
As we have seen, the above analysis leaves room for EDCs to be headed
by other degree morphemes than pos.8 For so, as, how, and too, this seems
unproblematic. However, what about meas? Evaluative modifiers are incompatible with measure phrases, and hence incompatible with meas, a property
they share with ordinary degree modifiers like very.
In section 1.2, I suggested that - at least in the case of very (and slightly,
pretty, fairly...) - the problem may be the requirement, imposed by meas on its
complement, that its degrees be measurable. Thus, tall can occur with meas
because degrees of height are measurable, but very tall cannot, because the
degree predicate returned by applying tall to very does not involve measurable
degrees. However, it is unclear how evaluative modification, which at no point
in its derivation involves a similar function from degree predicates to degree
predicates, could have the same scale-altering effect.
An alternative solution is provided in the syntactic literature: Corver (1997a,
2009) suggests that measure phrases, while overtly appearing in SpecDegP, originate in the same position as modifiers (SpecQP) and hence are incompatible
with them.9
I will not explore this issue further here; the important thing is that the
incompatibility of modifiers and measure phrases is a general problem, and not
a problem for the current analysis of evaluative modification in particular.
2.4.2
Back to the denotation of gradable adjectives
We may wonder whether an analysis similar to the one presented in this section
would be possible under the assumption that gradable adjectives denote measure
functions rather than degree predicates. The short answer is no. The somewhat
longer answer is that quantification over a degree argument is the only way to
give the degree scope over the propositional modifier.
(39)
DegP
Deg0
Deg
QPhe,di
λGλx[G(x) ≥ sG ]
Q0
AdvP
λp[surprising(p)]
Q0
λGλRλx[...?...]
APhe,di
λx[tall(x)]
8 Incidentally, the ability of EDCs to be headed by degree morphemes like so and how is
another thing that cannot be explained in Morzycki’s (2004) analysis.
9 This solution was pointed out to me by Norbert Corver (p.c.).
38
Unlike ˆtall(d)(x), tall(x) does not denote a proposition, so it is unclear
what eval could do to link the AP and the adverbial modifier in the first
place. Semantically, the adverb cannot do anything until much further into the
derivation, but that means it will always take scope over the degree construction.
The only possibility for a measure function-based approach, as far as I can
tell, is to reject the assumption that degree-modifying adverbs denote propositional modifiers. The fact that sentences like (40) are grammatical might
speak in the favour of an analysis in which the adverbs directly modify measure
functions:
(40)
Vernon’s height is remarkable.
One might wonder what exactly is going on here, though, as in the measure
function approach height(v) denotes a degree, and semantically, there is nothing remarkable about a degree as such. As Morzycki (2004) argues: “if a degree
is simply an interval on a scale as assumed here (...), predicating of this interval that it is remarkable or surprisingly or disappointing or strange would at a
minimum fail to make obvious predictions, and at worst might be irredeemably
incoherent”. For example, we may well assume that 7 is a degree. But to predicate of 7 that it is remarkable is meaningless. Adding a unit of measurement
hardly improves this: there is also nothing remarkable about 7 feet or 7 hours.
Now, having a height of 7 feet might be remarkable for an individual who is not
a flagpole or a giraffe, but ‘having a height of 7 feet’ can hardly be construed
as a degree anymore.
Concludingly, it might be hard, if not impossible, to analyse EDCs within a
measure phrase-based approach to degree constructions; and if it is possible, it
will not look anything like the analysis presented in this chapter.
2.5
Conclusion
In this chapter, I have critically examined the analysis of EDCs as developed
in Morzycki (2004) and found that it could be improved semantically as well
as syntactically. I have shown that there are several conceptual and empirical
problems with an analysis of EDCs as embedded exclamatives, and subsequently,
that the alternative Morzycki rejects is in fact perfectly valid if we assume that
gradable adjectives and evaluatives are monotone.
Furthermore, I have shown that the syntax for EDCs proposed by Morzycki
does not fit the data. My alternative consists of two major changes: first,
relocation of r from Deg0 to Q0 (and the evaluative AdvP to SpecQP), and
second, separation of r into eval and pos. This allows the EDC to be headed
by pos or some overt degree head, paralleling other degree constructions.
It also allows words like very and pretty to be treated syntactically like degree modifiers (occupying SpecQP), while still explaining why they occasionally
behave differently from evaluatives that occupy the same position: very, pretty
and other ‘true’ degree modifiers, being of type hhd, he, tii, hd, he, tiii (cf. Rett
39
2008), can directly modify the adjective without needing the intervention of
something like r.
One aspect of Morzycki’s analysis that stays the same in the present one is
the assumption that the denotations of evaluative adverbs and their adjectival
counterparts are identical. This will provide us with an excellent starting point
for the next chapter, in which we will look at EDCs involving gradable nouns
modified by adjectives.
40
Chapter 3
Degree modif ication of
gradable nouns
Although nearly all the literature on gradability and degree as summarised in
Chapter 1 refers only to adjectives, gradability, as we have seen, is not a property
of adjectives alone. So, with the syntax and semantics of adjectival modification
in place, it is now time to extend this analysis to modification of gradable nouns:
(1)
a. Vernon is an unbelievable idiot.
b. Maxwell is a horrible geek.
First, however, it is important to have a clear notion of what exactly constitutes gradability in nouns. Do gradable nouns have a degree argument, and do
they come with full-fledged degree scales, like adjectives? Or - another possibility - is their apparent gradability really some kind of metaphor or metalinguistic
process? In section 3.1, I will examine the options, and argue that gradable
nouns linguistically parallel gradable adjectives. Subsequently, section 3.2 will
be devoted to degree constructions involving nouns rather than adjectives, starting with a review of Morzycki’s (2009) analysis of degree modification by means
of size adjectives, and then combining insights from Morzycki and the previous
chapter into an analysis of EDCs involving nouns - sentences like Vernon is an
unbelievable idiot. Section 3.3 concludes this chapter.
3.1
What makes a noun gradable?
In an attempt to define the kind of noun that receives a degree reading when
modified by a size adjective like big, Morzycki (2009) writes:
Another way of articulating what I intend by ‘gradable noun’,
then, is that gradable nouns are those for which a single criterion
can be distinguished from the others as the most salient. For idiot,
41
it is stupidity (and not, say, animacy); for smoker, it is generally
frequency of or affinity for smoking (and not, say, ability to inhale
smoke); for goat cheese enthusiast, it is enthusiasm for goat cheese.
It is this ability to identify a single scale that distinguishes nouns
that admit degree readings of size adjectives from those that don’t.
While Morzycki’s description is definitely a step in the right direction, I feel it
is still not entirely accurate. After all, why wouldn’t chair be gradable, on a
single, one-dimensional scale of how much it resembles a typical chair? That is,
after all, exactly what we mean when we say things like This object is more a
chair than a table or This object is almost a chair. People would easily be able
to order a collection of more or less chair-like objects along such a scale. And
yet, we cannot refer to an extremely chair-like object as an unbelievable chair or
a big chair (in the interpretation ‘very chair-like’). Conversely, there are nouns
like nerd that are clearly gradable (as (1) shows), yet do not seem to “identify
a single scale”.
Sassoon (2007:2) proposes to treat all nouns as gradable, although her approach to gradability is different from that of Morzycki: in Sassoon’s view,
nouns have to be inherently gradable because their referents can be more or less
typical examples of the concept the noun denotes. That typicality is psychologically real has been demonstrated in many studies in the 1970s (for an overview,
see Rosch 1978 or Sassoon 2007). Rosch (1973 and subsequent research) experimentally confirmed Wittgenstein’s (1953) idea that category membership is not
a matter of meeting some definitional list of necessary and sufficient conditions;
rather, entities can be more or less representative of a certain category based
on their resemblance - across various dimensions - to an (abstract) prototypical
category member. For example, some dimensions usually associated with the
concept bird are having feathers, flying, nesting, singing, eating insects, and
small size. A bird that scores high on all these dimensions (say, a robin) will
therefore be considered a more ‘birdier’ bird than a huge bird that cannot sing
or fly (say, an ostrich).
Indeed, people are remarkably consistent in ordering entities along such a
typicality scale. Rosch (1973) also found that subjects’ reaction times in deciding whether an entity belonged to a certain category depended on that entity’s
typicality: the less typical the entity, the more time it takes for people to confim
their category membership.
It will be clear that if one defines gradability as bearing more or less resemblance to a prototype, a noun like chair is just as gradable as a noun like idiot:
just as there are various degrees to which one can be an idiot, there are also
various degrees to which an object can exemplify a chair. Furthermore, Sassoon claims that this conceptual gradability of nouns is reflected syntactically
in constructions like the following:
(2)
a. This is more a chair than a table.
b. This is almost/nearly a chair.1
1 There
is another reading in which some object or pile of building material is currently in
42
c. This is pretty/very much a chair.
However, as already noted above, there are also degree constructions in which
nouns like chair can not appear, whereas nouns like idiot can:
(3)
a. Vernon is a huge idiot.
∼The extent of Vernon’s idiocy is huge
b. This is a huge chair.
6∼The extent of this object’s chairness is huge
The differences between the above approaches raises the question what exactly
we mean when we are talking about the ‘gradability’ of nouns. Are we able to
interpret nouns in degree constructions as graded because of their conceptual
structure (in other words, is this a kind of metalinguistic gradability), or is
their gradability a truly linguistic property because nouns, like adjectives, have
a degree argument and come with a degree scale? And if so, what kind of degree
scale is this? Do all nouns have them? Why can we talk about an unbelievable
idiot but not in a similar way about an unbelievable chair ? And conversely, why
can we claim that an object is almost a chair while it is odd to say of a person
that he is almost an idiot?
3.1.1
Linguistic versus conceptual gradability
Let’s take a closer look at some of the data above. First of all, I would like to
note (2a) fits all the characteristics of metalinguistic comparison (cf. McCawley
1988, Morzycki 2008, Giannakidou & Stavrou 2009, Giannakidou & Yoon 2009),
which can also occur with non-gradable predicates. Therefore, this particular
construction does not provide evidence for the linguistic (rather than merely
conceptual) gradability of nouns.
The semantic difference between metalinguistic comparison and ordinary
comparative structures is exemplified in (4):
(4)
a. Snakes are more dangerous than spiders.
∼The degree to which snakes are dangerous is higher than the degree
to which spiders are dangerous
b. Spiders are more scary than dangerous.
∼It is more appropriate to call spiders scary than to call them dangerous.
There are several structural differences between ordinary and metalinguistic
comparison, some of which can be applied to comparisons involving nouns. First,
as McCawley (1988) observes, the place of the comparative morpheme (more)
may differ:
the process of transforming into a chair, and almost/nearly indicates that the endpoint of this
process has almost been reached. This is not the intended reading here.
43
(5)
a. Speeding is more legal than murder.
b. *Speeding is legal more than murder.
(6)
a. George is dumb more than crazy.
b. I am a machine now more than a man.
c. This object is a chair more than a table.
So, while both orders x more than y and more x than y can be used with metalinguistic comparison, only the latter is grammatical in ordinary comparatives.2
Secondly, some languages (e.g. Greek, Giannakidou & Stavrou 2009; Korean, Giannakidou & Yoon 2009) morphologically distinguish ordinary from
metalinguistic comparatives. A similar phenomenon can be found in English:
in metalinguistic comparatives, it is possible to use the comparative morpheme
rather (Dutch: eerder ) instead of more, while this is ungrammatical in ordinary
comparatives:
(7)
a. George is rather dumb than crazy.
b. This object is rather a chair than a table.
c. *George is rather dumb than Bill.3
And crucially, metalinguistic comparison is possible with predicates that are not
normally gradable:
(8)
a. Your problems are more financial than legal.
(McCawley 1988)
b. *Your problems are more financial than mine.
The above shows that the appearance of nouns in seemingly degree-related constructions does not necessarily provide evidence for their gradability. Metalinguistic comparison, in fact, is an excellent example of how certain constructions are interpretable by virtue of their conceptual structure, without requiring
this structure to have a linguistic counterpart. A sentence like (9), for example, shows that metalinguistic comparison between predicates denoting ‘crisp’,
sharply delimited concepts is uninterpretable:
(9)
#This number is more odd than even.
Likewise, we would not be able to make sense of a sentence like (6b) if it were
not conceptually possible for a machine to be partly not a machine, or for a
2 But
note that these sentences are not exactly comparable. In (5), legal is predicated of
both speeding and murder. In (6a), however, it is not the case that dumb is predicated of both
George and crazy (‘George is dumb more than crazy is dumb’). So it’s not just the place of the
comparative morpheme that’s different - it’s the whole construction. The proper equivalent of
(5) would be George is dumb more than he is crazy. Luckily, this is also grammatical, unlike
*? Speeding is legal more than murder is, so the gist of McCawley’s argument stays the same.
3 This sentence can be made to work with contrastive focus intonation on George and Bill,
but note that the result is again a case of metalinguistic comparison, paraphrasable as ‘It is
more appropriate to call George dumb than it is to call Bill dumb.’
44
man to have just enough manly properties for the use of the word man to be
still appropriate.
In short, we have to distinguish carefully between actual linguistic gradability
(the property of having a degree argument) and mere conceptual gradability.
But the above does not mean that all seeming gradability in nouns can be
reduced to a combination of a certain conceptual structure and metalinguistic
operations. In the remainder of this section, I will show that gradable nouns, like
adjectives, behave like monotone predicates, providing evidence for the claim
that they are in fact linguistically gradable and come with a degree argument.
My eventual proposal on what constitutes gradebility in nouns will contain
elements of both Morzycki’s and Sassoon’s approaches.
3.1.2
Monotonicity
In the previous chapter (section 2.2.2), we have seen that the correct interpretation of EDCs can be ensured by the fact that degree predicates and evaluative
adverbs are monotone (Heim 2000, Katz 2005, Nouwen 2005). Thus, claiming
that it is unbelievable for Maxwell to have some degree of tallness d is only valid
if Maxwell’s having any degree of tallness d0 ≥ d would also be unbelievable.
This means, for example, that we cannot use (10a) in the context of (10b), even
if being unbelievably short technically equals having an unbelievable degree of
tallness (cf. (18) in section 2.2.2):
(10)
a. ‘Maxwell is unbelievably tall.’
b. Maxwell is, in fact, unbelievably short.
Monotonicity can also explain Zwicky’s (1970) observation that in adverb
pairs like usually/unusually, the negative adverb can function as a degree modifier
while its positive counterpart can only be interpreted as a sentential adverb
(Nouwen 2009):
(11)
a. The children are unusually noisy.
b. The children are usually noisy.
Unusually (like remarkably) being downward monotone, (11a) implies that it
would also have been unusual for the children to be even more noisy than they
are now; however, it is well possible that there is some degree of noisiness
d0 < d that is not unusual for the children at all. Conversely, usually is upward
monotone, which preserves rather than reverses entailments:
(12)
a. O is MON↑ iff p |= p0 ⇒ O(p) |= O(p0 )
b. Mildred is reading a Booker Prize-winning novel |= Mildred is reading
a novel
c. It is usual for Mildred to be reading a Booker Prize-winning novel |=
It is usual for Mildred to be reading a novel
45
So, what exactly does it mean to be ‘usually noisy’ ? Recall that the monotonicity of noisy implies that if x is noisy to some degree d, it is noisy to all degrees
d0 < d as well. Therefore, every x is noisy to the lowest possible degree (let’s
call it d0 ). This in turn means that if there is any degree d of noisiness for which
it holds that usual(noisy(d)(x)), it also holds that usual(noisy(d0 )(x)). But
noisy(d0 )(x) is true of every x. Therefore, any x that is noisy is also usually
noisy - the degree reading of x is usually noisy is trivially true. This presumably
explains why it is not attested.
What is interesting for our purpose here is that the above observations also
hold for nouns:
(13)
a. Vernon is an unbelievable genius.
b. Vernon is a believable genius.
First, (13a) cannot be used to describe a situation in which Vernon is in fact
extremely stupid, even though it is probably quite unbelievable that someone
could have an IQ like Vernon’s. Secondly, while (13a) has a degree reading,
(13b) has not. It is quite imaginable that we could interpret (13b) as something
like ‘Vernon bears an average resemblance to the prototypical genius’, and yet
we don’t.
Nouwen’s monotonicity analysis - or any analysis that explains the same
set of facts - could easily deal with the data in (13), provided that we assume
that nouns are degree predicates (as monotonicity is defined as a property of
predicates of type hd, eti (Heim 2000)).4
3.1.3
Scales and prototypes
Let us briefly return to Sassoon’s (2007) notion of nominal gradability, which
is based on typicality. This is why we can easily interpret constructions like
almost a chair or pretty much a chair - what we mean is ‘almost a typical chair’
or ‘pretty much a typical chair’. The typical chair itself, the prototype, is the
epitome of ‘chairness’ - one simply cannot get more chair-like than that.
Translating this into other familiar terms, if nouns like chair were associated
with a scale, its maximum would be represented by a complete similarity to the
prototype. In other words, the scale of chair (or any other noun denoting
a concept with a prototype) would be closed on the upper end (cf. section
1.1.4, Kennedy & McNally 2005). One property of closed degree scales is that
they allow endpoint modifiers like almost, pretty much, and completely, and
disallow intensifiers like very, enormously, and terribly; this fits with our earlier
observations on nouns like chair, which allow almost but lack anything like a
degree reading when modified by an evaluative or a size adjective.
(14)
Closed scale: open, empty, invisible
4 The fact that gradable nouns can be modified by size adjectives (e.g. big idiot) provides
even more evidence for their having a degree argument, as we will see when we discuss the
analysis of size adjectival modification later in this chapter.
46
a. Maxwell’s pet elephant is almost/completely/virtually invisible.
b. *Maxwell’s pet elephant is unbelievably/hugely/terribly invisible.
On the other hand, open-scale predicates show precisely the opposite pattern,
as exemplified in (15):
(15)
Open scale: tall, expensive, stupid
a. *Vernon is almost/completely/virtually stupid.
b. Vernon is unbelievably/hugely/terribly stupid.
As we have seen, the nouns discussed by Morzycki (idiot, nerd, soccer fan,
airhead, goat cheese enthusiast, simpleton, Barbie doll lover, loser, weirdo...)
pattern with these open-scale adjectives:
(16)
a. *Vernon is almost/completely/virtually an idiot.5
b. Vernon is a(n) unbelievable/huge/terrible idiot.
We conclude that nouns like idiot should be associated with an open rather
than a closed scale - but from this, it follows (tracing the steps of our argumentation backwards) that the concepts they denote cannot have a prototype. This
corresponds to our intuition about concepts like bird and chair versus concepts
like nerd, idiot and goat cheese enthusiast: while we can quite easily define a
prototype for chair - the most chair-like kitchen chair we can imagine - it is
much harder to come up with a prototypical nerd. In fact, we will find that we
can always come up with a nerd that is more nerdy than our putative prototype
- someone whose IQ is ever so slightly higher, whose eyesight is just the tiniest
bit worse, whose programming skills extend to just one more programming language. This means that perhaps the most crucial property of gradable nouns
may be defined in terms of prototypes, or rather, lack thereof: gradable nouns
denote concepts that lack a prototype.
Of course, this definition is incomplete as long as we cannot tell whether
the reverse is also true: are all nouns that denote concepts without a prototype
gradable?
First, note that there does seem to be a way to think of the concepts denoted
by gradable nouns in terms of prototypicality. Take genius. There are many dimensions one can associate with genius - high IQ, natural talent for a particular
art or science, excellence at a very young age, representing a turning point in
5 In some cases, the judgements are less clear; for example, both virtually a sociopath and
an unbelievable sociopath are perfectly fine. But note that the sense in which sociopath is
used in these constructions differs: in the first, a sociopath is defined in psychiatric terms as
someone who meets a set of conditions (idiot used to have a similar use, in which case almost
an idiot would have been perfectly acceptable as well). One can almost meet a certain set of
conditions, hence one can be almost a sociopath. In the case of an unbelievable sociopath and
the contemporary use of idiot, however, the term is used much more loosely, and can apply to
anyone who, to some extent, possesses the characteristics we associate with sociopathy. It is
as inappropriate to call a sociopath in the former, diagnostic sense unbelievable or huge as it
is to say of a door that it is unbelievably open, or of someone paralysed to the neck that she
is enormously lame or a huge cripple (which would be rude in addition to ungrammatical).
47
Figure 3.1: A prototypical genius?
the history of their field - and no matter which genius we pick, we can come up
with someone who embodies these dimensions more. However, it should not be
surprising if some particular geniuses spring more readily to mind than others
(fig. 3.1), in a way that is hardly predictable from their ranking among their
fellow geniuses.
This is reminiscent of a series of studies by Armstrong et al. (1983), who
show that conceptual structure and exemplariness are two separate things.
When asked to rate different even numbers for exemplariness of the even number
category, subjects gave graded responses (favouring 2 and 4 over 34 and 806, for
example), even though they all agreed that no even number could be ‘more’ or
‘less’ even than any other even number. In other words, category exemplariness
may be graded, even if category membership is not. (So while Einstein may be
an exemplary genius, this does not necessarily make him prototypical.)
The concept even number is an example of a definitional concept, which
lacks a prototype. It is not gradable, so our above definition is not reversible.
However, there is a clear difference between genius and even number : category
membership is graded for the former, but clear-cut for the latter.
Thus, we have now identified two dimensions along which to classify the concepts denoted by nouns (existence of a prototype and gradedness of category
membership). The first class is exemplified by nouns like chair, whose category
membership is graded and which have a prototype. The second class is exemplified by nouns like even number, which have neither. The third class are our
gradable nouns, whose category membership is graded but which do not have
a prototype; this is the only class that we can demonstrate to be linguistically
gradable (by monotonicity). The fourth logical possibility, concepts with clearcut membership as well as a prototype, does not exist, as prototypicality only
applies to categories with graded membership.
48
chair
nerd
even number
prototype?
y
n
n
graded membership?
y
y
n
ling. gradable?
?
y
n
We can now refine our proposed definition of ‘gradable noun’: a noun will
be linguistically gradable iff it denotes a concept that a) is conceptually gradable
(i.e. has graded membership), and b) lacks a prototype. This definition assumes
that Sassoon’s ‘gradable nouns’ - nouns like bird and chair - are, in fact, not
linguistically gradable, in spite of their ability to occur with modifiers like almost and virtually. As far as I can tell, there is no decisive way to show that we
are not dealing with some metalinguistic process here. But if one would wish to
leave open the option that chair is also linguistically gradable, one might read
the above definition as simply delimiting the kind of gradable noun I want to
talk about in this thesis, and which I will henceforth always intend when talking about a ‘gradable noun’ - the ‘open-scale’ class of idiot, nerd, soccer fan,
airhead, goat cheese enthusiast, simpleton, Barbie doll lover, loser, and weirdo.6
There is one final scale-related issue that I should mention before going on
to the next section. Above, I suggested that gradable nouns necessarily map
unto open scales, which is why they are incompatible with endpoint modifiers.
Interestingly, though, Morzycki (2009) distinguishes between open- and closedscale nouns within the class of gradable nouns, based on their ability to occur
with complete: a complete idiot/nerd is possible, but a complete goat cheese
enthusiast is not. So, does that mean that the idiot scale has an endpoint after
all?
I would like to argue that the difference between idiot/nerd and enthusiast
is in fact another one. Complete or total in a complete/total nerd do not really
seem to refer to the endpoint of a scale, such that a total nerd could not possibly
be more nerdy; rather, they carry a sense of having some property ‘in every
respect’. Thus, a total nerd is someone who is nerdy with respect to his looks,
social skills, intelligence, hobbies, in short, every dimension that we tend to
associate with nerdiness. In contrast, whether someone can reasonably be called
a goat cheese enthusiast or a curling fan really only depends on one dimension:
their enthusiasm for goat’s cheese and curling. This analysis correctly predicts
that nouns that can occur with completely also allow modification by a with
respect to clause, while the opposite holds for nouns that cannot occur with
completely:
(17)
a. Maxwell is a nerd with respect to his computer skills.
b. Vernon is an idiot with respect to his political views.
6 As
far as I can tell, mass nouns derived from gradable adjectives - wisdom, generosity;
cf. Nicolas (2004) - also fall under the definition. Indeed, they can be modified by evaluatives
and size adjectives, resulting in a degree reading; this is impossible for ‘ordinary’ mass nouns
like beer and furniture. I will not concern myself with mass nouns in this thesis (admittedly,
because I thought of the issue too late to be able to study them properly) but the reader
might want to bear them in mind.
49
c. *Mildred is a goat cheese enthusiast with respect to her fan club
attendance.
d. *Herman is a Barbie doll lover with respect to the size of his collection.
The claim that nerd and idiot are not associated with a closed scale is further
supported by the difference between the dialogues in (18a) and (18b):
(18)
a. ‘My grandmother went totally blind last year.’
- #‘That’s horrible. Do the doctors expect her to get even blinder?’
a total nerd
b. ‘I can’t believe you’re dating Maxwell. He’s
.’
totally nerdy
- ‘Look who’s talking. Your boyfriend is even nerdier.’
If the total(ly) in (18b) were really the same as the totally in (18a), namely an
endpoint modifier, the response in (b) would be just as infelicitous as that in
(a). The fact that it is not shows that the two total s do not have the same
function, and this in turn means that the fact that we can talk about someone
being totally nerdy does not imply that the scale of nerdiness is actually closed.
But this also means that total(ly) in a total nerd or totally nerdy is not actually
a degree modifier - there is nothing ‘total’ or ‘complete’ about said individual’s
degree of nerdiness. Rather, it modifies dimension.
3.2
Syntax and semantics of noun modif ication
In the previous section, we have concluded that gradable nouns are linguistically
gradable - i.e. denote degree predicates - and that they are associated with open
degree scales. We have also seen that they can be modified by much the same
elements as gradable adjectives. In this section, we wil look more closely at
degree modification of nouns. Section 3.2.1 discusses degree modification by size
adjectives, first presenting the analysis of Morzycki (2009) and then restating
this analysis in terms of degree predicates rather than measure functions; once
again, we will see the importance of monotonicity. Section 3.2.2 discusses degree
modification by evaluative adjectives.
3.2.1
Degree modif ication by size adjectives
One (apparently universal) property of gradable nouns, as identified by Morzycki (2009), is that they can be modified by size adjectives like big, huge, and
enormous, as in (19-19):
(19)
(20)
Vernon is a huge idiot.
a. Herman is een enorme idioot.
‘Herman is an enormous idiot’
50
(Dutch)
b. Janez je velik ljubitelj živali.
John is big lover animals.GEN
‘John is a big animal-lover’
(Slovene)
c. Ani ma’aritz gadol shel habolshoi.
I admirer big of the.bolshoi
‘I am a great admirer of The Bolshoi’
(Hebrew)
d. John-wa oo-bakamono desu.
John-TOP big-fool is
‘John is a big fool’
(Japanese)
It is not just certain well-established size adjectives that can modify degree,
but any adjective - including neologisms made up for the occasion - as long as
they predicate bigness. Adjectives that predicate smallness fail to result in a
degree reading (following Morzycki, I use % to indicate that the degree reading
is absent):



 small
tiny
idiot.
(21) %Vernon is a


diminutive
Morzycki calls this the ‘Bigness Generalisation’. Another generalisation that
holds crosslinguistically is the ‘Position Generalisation’, which states that degree
readings of size adjectives are possible only in attributive (and not in predicative
or postnominal) positions:
(22)
(23)
%That idiot is big.
a. Un gran idiota
‘a great idiot’
(Spanish)
b. %Un idiota grande
‘an idiot who is physically large’
Morzycki argues that size adjectives operate on degrees the same way measure
phrases do: they both “predicate of a degree that it has a certain minimum
size”. A measure phrase like six feet, in this view, takes a degree d and returns
a truth value - in this case, 1 for all degrees equal to or greater than six feet,
and 0 for all smaller degrees. Similarly, a size adjective like big takes a degree d
and returns 1 if d is ‘big’ and 0 if it is not. This means that size adjectives like
big are abstract enough to be able to measure the size of degrees themselves which in turn means they must be flexible between functions of type he, di and
type hd, di. The latter, when applied to a similarly flexible form of pos (size
adjectives, after all, are gradable and hence require degree morphology for their
interpretation), results in a measure phrase of type hd, ti - the same type as an
‘ordinary’ measure phrase like six feet.7
7 In Chapter 1, we have not really bothered with the syntactosemantic details of measure
phrases, following Kennedy (1997) and Svenonius & Kennedy (2006) in treating them simply
51
In line with this, it seems reasonable to propose a DegP for nominals that
parallels adjectival constructions with a measure phrase and a degree head meas.
However, there is an important semantic difference between Vernon is six feet
tall and Vernon is a big idiot: the former does not entail the corresponding
positive form, but the latter does. Morzycki remedies this by adjusting the
denotation of measN , the nominal counterpart of meas, to include the function
of pos:
(24)
[[measN ]] = λGλM λx[min{d : d ∈ scale(G)∧M (d)} ≤ G(x)∧sG ≤ G(x)]
(where M is a measure phrase consisting of a size adjective, scale(G)
is the scale (ordered set of degrees) that comes with a particular gradable
noun G, and min picks out the smallest degree on that scale that satisfies
M)
For example, for Vernon is a big idiot to be true, Vernon’s idiocy should be
greater than or equal to the smallest degree on the idiot scale that can still be
called ‘big’, and greater than or equal to the idiot standard. (25) shows how
the denotation of big idiot is computed:
(25)
a.
DegN Phe,ti
DegN 0hdt,eti
DegPhd,ti
Deghdd,dti
APhd,di
DegN hed,hdt,etii
NPhe,di
pos
big
measN
idiot
b. [[pos]] = λGλd[sG ≤ G(d)]
[[pos big]] = λd[sbig ≤ big(d)]
[[measN idiot]] = λM λx[min{d : d ∈ scale(idiot)∧M (d)} ≤ idiot(x)
∧ sidiot ≤ idiot(x)]
like degrees. Thus, in Vernon is six feet tall, six feet denotes a particular degree of tallness
which Vernon’s height is stated to equal (or exceed). Morzycki, however, follows Schwarzschild
(2005) in assuming that measure phrases are functions from degrees to truth values - in other
words, predicates rather than arguments (Schwarzschild (2005) provides several arguments for
this). This means that his version of meas looks different, too (note that both Kennedy and
Morzycki treat gradable predicates as measure functions of type he, di, which is a somewhat
different approach from the one we have taken in the previous chapter):
(ii) [[meas]] = λAλdλx[d ≤ A(x)]
(Kennedy 1997)
(iii) [[meas]] = λAλM λx[min{d : M (d)} ≤ A(x)]
(Morzycki 2009)
(where M is a measure phrase and min{d : M (d)} yields the smallest degree that
satisfies the measure phrase)
It should be clear that (ii) rules out the possibility of a direct parallel between measure phrases
and size adjectives; there is no way we can analyse big and enormous as denoting degrees.
52
[[[pos big] measN idiot]] = λx[min{d : d ∈ scale(idiot) ∧ sbig ≤
big(d)} ≤ idiot(x) ∧ sidiot ≤ idiot(x)]
How does this analysis account for the Position and Bigness Generalisations?
The former is rather trivial: degree readings of size adjectives can be computed
only from the above syntactic configuration. The latter follows from the present
analysis in a way reminiscent of the monotonicity-related arguments we have
seen earlier.
Consider the denotation of small idiot, computed along the same lines as
(25b) above:
(26)
[[[pos small ] measN idiot]] = λx[min{d : d ∈ scale(idiot) ∧ ssmall ≤
small(d)} ≤ idiot(x) ∧ sidiot ≤ idiot(x)]
Thus, for an individual x to be a small idiot, x’s degree of idiocy should equal
or exceed the smallest degree on the idiot scale (picked out by min) that satisfies the smallness standard ssmall . But the smallest degree on the idiot scale
(which presumeably satisfies the smallness standard, or no degree would) is the
same in every case: d0 , the minimum of the idiot scale, which corresponds to a
complete lack of idiocy. This means that any degree of idiocy equals or exceeds
this - in other words, x is a small idiot is trivially true whenever x is an idiot
is true. Because the size adjective does not make any semantic contribution
on a degree reading, the rules of pragmatics dictate that we should choose the
reading on which it does, and interpret small idiot as a literal assertion about
said idiot’s physical size.8 It can easily be verified that this is a property of all
size adjectives predicating smallness.
One question remains. In the previous chapter, we have based our analysis
of EDCs on the assumption that gradable adjectives denote functions from degrees to properties; the alternative, in which gradable adjectives denote measure
functions, was shown not to work. My analysis of nominal EDCs (in the next
section) will therefore again be based on hd, he, tii-type predicates. Obviously, a
uniform analysis of size and evaluative adjectival modification would be preferable over one that has size adjectives modifying measure functions and evaluative adjectives modifying degree predicates. We may therefore wonder whether
it is possible to convert the analysis of degree modification by size adjectives,
as outlined above, to a degree predicate-based approach.
In fact, this conversion is pretty straightforward. First, let’s see what the
type and denotation of meas should be in a hd, he, tii-type approach. Arguably,
it should largely parallel the semantics of pos - the principal difference between
pos and meas being that pos compares the degree to which an individual x has
some property G to the standard of G, while meas compares it to an overtly
expressed measure. (27) shows the denotation of pos:
8 Alternatively, we may want to claim that the idiot scale, like for example the scales of
height, weight and age, does not have a minimum, but asymptotically approaches zero (cf.
section 1.1.4). In that case, the computation of the value of [[small idiot]] would fail because
the min operator would be unable to return a degree.
53
(27)
[[pos]]hhd,eti,hetii = λGλx∃d[G(d)(x) ∧ d ≥ sG ]
In the case of meas, the comparison value (which we will call d0 ) is not computed
on the spot based on contextual information (like sG ), but is supplied later in
the derivation, by a measure phrase. This means we need an additional lambdaabstraction over d0 , and a corresponding type adjustment:
(28)
[[meas]]hhd,eti,hd,etii = λGλd0 λx∃d[G(d)(x) ∧ d ≥ d0 ]
However, note that this denotation of meas does not allow us to draw Morzycki’s
parallel between size adjectives and measure phrases, as it calls for an argument
of type d - something that, unless we propose some exotic function from size
adjectives to degrees (but what would something like that even mean, without
a particular scale to refer to?), a size adjective simply cannot be or become.
One alternative would be to raise the type of the measure phrase from d to
hhd, eti, hetii, but it is still unclear how we could arrive at such a type (starting
out with a size adjective of type hd, he, tii) without hugely complicating things.
So we will adopt Morzycki’s suggestion, based on Schwarzschild (2005), that
measure phrases are functions from degrees to truth values (exactly how to
derive this type from a numeral or quantifier phrase like six feet or several
years remains an open question). The denotation of meas now looks like the
following:
(29)
[[meas]]hhd,eti,hdt,etii = λGλM λx∃d[G(d)(x) ∧ M (d)]
(where M is a measure phrase and G a gradable adjective)
This denotation lacks the min operator we saw earlier, but as we will see, its
effect - of making x is a small idiot and the like trivially true - will here be taken
over by the monotonicity of the size adjective. Neither does it need an explicit
equivalent of the requirement that d ∈ scale(G): in the present approach, this
follows naturally from the fact that d is an argument of the predicate G.
As before, [[measN ]], which is used with nominal degree predicates, needs to
entail the positive form:
(30)
[[measN ]]hhd,eti,hdt,etii = λGλM λx∃d[G(d)(x) ∧ d ≥ sG ∧ M (d)]
(where M is a graded size adjective of type hd, ti, and G a gradable noun)
Here, then, is how the semantics of big idiot is derived in the translated
version of Morzycki’s analysis:
(31)
a.
DegN Phe,ti
DegN 0hdt,eti
DegPhd,ti
Deghhd,eti,dti
APhd,eti
DegN hhd,eti,hdt,etii
NPhd,eti
pos
big
measN
idiot
54
b. [[pos big]] = λx∃d0 [big(d0 )(x) ∧ d0 ≥ sbig ]
[[measN idiot]] = λM λx∃d[idiot(d)(x) ∧ d ≥ sidiot ∧ M (d)]
[[[pos big] measN idiot]] = λx∃d[idiot(d)(x)∧d ≥ sidiot ∧∃d0 [big(d0 )(d)∧
d0 ≥ sbig ]]
The Position Generalisation is accounted for in the same way as before. The
Bigness Generalisation, however, now follows from the monotonicity of the size
adjective. Recall that by our definition of monotonicity, if an individual x has
some property to a degree d, it also has this property to all lower degrees d0 ≤ d.
This means that x is d-big is true for every x whose degree of bigness equals
or exceeds d. Similarly, x is d-small is true for every x as small as, or smaller
than, d indicates. (32) shows the denotation of x is a small idiot:
(32)
[[[pos small] measN idiot]] = λx∃d[idiot(d)(x)∧d ≥ sidiot ∧∃d0 [small(d0 )(d)∧
d0 ≥ ssmall ]]
This is true if (a) x is an idiot, and (b) the degree d to which x is an idiot is
d0 -small. The latter is true if there is some degree d0 that satisfies the smallness
standard for which it holds that small(d0 )(d). But this means that our (b)requirement can be fulfilled by any d ≥ d0 . Because the smallness scale has a
maximum that asymptotically approaches zero (a size of zero means not having
a size at all, so it cannot be included on any linguistic scale that measures size),
every degree on the smallness scale is equalled or exceeded by some other degree.
So no matter how high up on the smallness scale d0 is, there will always be a
higher d available.
Now, let’s assume that d is infinitely high - which, in this case, corresponds
to infinite smallness. We can now prove that this degree - which must necessarily
satisfy our (b)-requirement - exists whenever ∃d[idiot(d)(x)∧d ≥ sidiot ] is true,
and thus that (b) is trivially satisfied. To see why this is so, consider the fact
that idiot is monotone as well. Therefore, every x that is idiotic to some degree
d00 , is also idiotic to all lower degrees d0 , up to the inf initely smallest degree d.
This means that idiot(d)(x) is, indeed, always true. Just as before, x is a small
idiot is true whenever x is an idiot is true, which makes the addition of small
semantically vacuous on a degree reading.
In short, the Bigness Generalisation holds because both the gradable predicate and the size adjective are monotone. Again, just as before, the above proof
can be generalised to all size adjectives M whose scale is ordered such that
degrees high on the G-scale correspond to degrees low on the M -scale, which
is true for any adjective that predicates smallness: more idiocy, for example,
obviously means a less small degree of idiocy.
To summarise, degree modification by size adjectives can be analysed on a
par with measure phrase modification. To make this work, we have to assume
both measure phrases and graded size adjectives are functions from degrees to
truth values (Schwarzschild 2005, Morzycki 2009, Corver 2009). The degree
55
morpheme at work, in both cases, is meas; when gradable nouns rather than
adjectives are concerned, meas needs to incorporate the denotation of pos as
well.
The approach can be formalised more or less equally well based on either
measure phrases (he, di) or degree predicates (hd, hetii). There are a few technical differences between the two analyses; one could argue that the measure
phrase analysis has to specify explicitly what follows more naturally from existing assumptions in a degree predicate analysis. In both cases, the Position
Generalisation is rather trivially accounted for by the fact that the semantic
computation necessary for a degree reading can only take place in a particular
syntactic configuration. The Bigness Generalisation is dealt with in slightly
distinct ways, but the idea behind them is essentially the same.
3.2.2
Degree modif ication by evaluative adjectives
Below, the syntax and semantics of Evaluative Degree Constructions (EDCs) as
argued for in Chapter 2 are repeated:
(33)
DegPhe,ti
Deg0
Deghhd,eti,eti
QPhd,eti
pos
Q0hhst,ti,hd,etii
AdvPhst,ti
R
Qhhd,eti,hhst,ti,hd,etiii
APhd,eti
eval
G
where R is an evaluative adverb, G a gradable adjective, and
(34)
[[eval]] = λGλRλdλx[G(d)(x) ∧ R(ˆG(d)(x))]
Our zero hypothesis is that the evaluative degree modification of nouns has
precisely the same structure, with the adjective in the Spec of QP and the NP
as the complement of eval. In principle, this could work fine: gradable nouns
have the same type as gradable adjectives, and, following Morzycki (2004), evaluative adjectives and adverbs have identical denotations. Any explanation of
the behaviour of evaluatives in combination with other elements can straightforwardly be extended to nominal EDCs. For example, the possibility of having
both a size adjective and an (evaluative) modifier is ruled out by the same requirement that rules out the modification of a gradable adjective by both an
evaluative and a measure phrase, as in (35b):
56
(35)
a. *Vernon is six feet incredibly tall.
b. %Vernon is a huge unbelievable idiot.
Even having an AP in the specifier position (of some functional projection on
top of NP) is not unusual: Cinque (1994) argues that this analysis of attributive
APs is to be preferred over one that treats them as adjuncts to DP or NP.9 This
also suggests that degree and non-degree readings of EDCs may have a largely
identical syntactic structure.
There is a potential problem here. The reader may recall that it was argued
in section 2.3.3, following Corver (1997b), that Dutch gender agreement between
adverb and adjective is only possible when these elements stand in a Spec-Head
relation to each other; when eval interferes between them, there is no gender
agreement. In line with this, we should expect a lack of gender agreement
between adjective and noun in Dutch nominal EDCs - semantically, eval has
to interfere between the adjective and the noun, or we would end up with a
type mismatch. Contrary to this expectation, however, adjective and noun
obligatorily agree in gender:
(36)
a. Een ongelofelijk-∅ kreng
‘an unbelievable-NEUT bitch’
b. Een ongelofelijk-e sukkel
‘an unbelievable-NONNEUT loser’
Trying to find a solution for this apparent contradiction requires venturing deep
into syntactic terrain, in particular more recent Minimalist literature on approaches to agreement like Move-F (cf. Chomsky 1995) or Agree (cf. Chomsky
2004, Schoorlemmer 2009), and the debates still surrounding it; see for example Corver (2006) for a more recent and syntactically much more complicated
account of adverb-adjective (‘proleptic’) agreement in Dutch adjectival DegPs.
I have decided not to go into this here, as it would likely double or triple the
length of this chapter. My suggestion, which may or may not hold up under further syntactic scrutiny, is that the agreement difference between adjectival and
nominal EDCs has something to do with the fact that valuation of the gender
feature is optional in adverbs, but obligatory in (attributive) adjectives; thus,
adjectives may force covert movement of the noun’s gender features (under a
Move-F approach) to some functional projection where Spec-Head agreement
can take place, while a similar operation in adjectival EDCs would be ruled out
by economy considerations.10
A more relevant problem with our putative DegN P is that there is no reason
why DegN 0 should select a QP complement. Recall that in the adjectival DegP,
Q0 may contain morphemes like more, less, -er and the dummy much. None of
these has a direct counterpart in the DegN P: comparatives involving gradable
9 Thanks to Coppe van Urk for bringing Cinque’s view to my attention, and also for linking
me to Corver (2006).
10 I am not sure how to formulate this in terms of the Probes and Goals of the more recent
Agree approach, but according to Hornstein et al. (2005:9), the empirical coverage of the two
approaches is the same.
57
nouns require a pseudopartitive, like more of an idiot and less of an elephant
lover.11 Furthermore, while certain degree modifiers, like rather and quite, can
combine with gradable nouns (Vernon is rather/quite an idiot), they are clearly
not located inside DegN P, as shown by the intervention of the determiner. This
suggests that the usual position for degree modifiers, SpecQP, is unavailable. In
short, it is reasonable to assume that the QP projection is absent from DegN P.
But this renders eval homeless, too.
Finally, we may wonder to what extent the degree reading and the intersective, non-degree reading of EDCs are based on the same syntactic structure.
Speaking purely in terms of linear order, pos-AP-eval-NP would result in a
degree reading, and AP-pos-NP in a non-degree reading (if we assume Heim &
Kratzer’s Predicate Modification rule (Heim & Kratzer 1998), an intersective
interpretation is only possible when both adjective and noun are of type he, ti;
this means that the noun must be ‘graded’ by degree morphology before it is
combined with the adjective). Do either pos, AP, or both occupy the same position here, or are we talking about two completely different structures? And, a
related issue, to what extent does the structure of EDCs parallel the structure of
degree modification by size adjectives? From a conceptual point of view, proposing a single DegN structure that can account for both kinds of modification and
both the degree and non-degree readings is to be preferred over having three or
four different analyses. Also, there is no syntactic evidence that they should be
structured differently; for instance, both degree and non-degree readings show
precisely the same agreement patterns in Dutch, and more complicated degree
constructions - like comparatives - show that evaluatives and size adjectives can
appear in precisely the same configurations:
bigger
(37)
a. Maxwell is a
animal lover than Vernon.
more unbelievable
big
b. Vernon is as
an animal lover as Maxwell is.
unbelievable
c. Maxwell is a hugely unbelievable animal lover.
d. Maxwell is an unbelievably huge animal lover.
The simplest way of analysing all the aforementioned forms of noun and degree modification on a par, in fact, immediately becomes clear when we match
their ingredients with the available positions inside DegN P. What is important
here is that there is no QP; our DegN directly selects the NP as its complement,
which leaves us with only only one available head position for degree morphology - Deg0 - and one available specifier position for the AP - SpecDegP. This
straightforwardly accounts for the intersective interpretations of both size and
evaluative modification:
11 These
quantifiers can be used with mass nouns, but that is a different situation.
58
(38)
DegN Phe,ti
AP
DegN 0he,ti
pos
NPhd,eti
(39) shows the semantic derivation of the non-degree reading of terrible idiot
(abstracting over the fact that terrible itself is a DegP). The literal size reading
of constructions like big idiot is derived in precisely the same way.
(39)
[[terrible pos idiot]]
= λGλx∃d[G(d)(x) ∧ d ≥ sG ]([[idiot)([[terrible]])
= λGλx∃d[G(d)(x) ∧ d ≥ sG ](λdλx[idiot(d)(x)])([[terrible]])
= λx∃d[idiot(d)(x) ∧ d ≥ sidiot ]([[terrible]])
= λx∃d[idiot(d)(x) ∧ d ≥ sidiot ](λx[terrible(x)])
= (by Predicate Modification) λx[terrible(x) ∧ ∃d[idiot(d)(x) ∧ d ≥
sidiot ]]
In line with our non-degree interpretation of terrible idiot, this asserts of some
individual x that x is both terrible and an idiot.
Note that this is in fact the exact same structure assumed in section 3.2.1 for
degree modification by size adjectives (with pos replaced by meas, obviously),
so we have that one covered as well. This leaves only the degree interpretation
of nominal EDCs. This one is less straightforward, as it supposedly contains two
heads (pos and eval), but there is only one head position available between AP
and NP. Adding a projection headed by pos on top of our DegN P would lead to
the right semantics, but conceptually there would be something not quite right
about it (cf. section 1.1.2) - why two DegPs on top of each other when we are
so clearly dealing with a single degree construction?
I therefore propose that evalN needs to incorporate the denotation of pos perhaps not coincidentally parallelling the difference between meas and measN
discussed earlier. This results in a Deg0N of type hhst, ti, hetii, rather than
he, ti, so an intersective interpretation via PM is impossible. Hence, the evaluative adjective is interpreted as a propositional modifier12 , enabling the correct
derivation:
(40)
a. [[evalN ]] = λGλRλx∃d[G(d)(x) ∧ d ≥ sG ∧ R(ˆG(d)(x))]
b. [[terrible eval idiot]]
= λGλRλx∃d[G(d)(x) ∧ d ≥ sG ∧ R(ˆG(d)(x))]([[idiot]])([[terrible]])
= λGλRλx∃d[G(d)(x)∧d ≥ sG ∧R(ˆG(d)(x))](λdλx[idiot(d)(x)])([[terrible]])
= λRλx∃d[idiot(d)(x) ∧ d ≥ sidiot ∧ R(ˆidiot(d)(x))]([[terrible]])
= λRλx∃d[idiot(d)(x)∧d ≥ sidiot ∧R(ˆidiot(d)(x))](λp[terrible(p)])
= λx∃d[idiot(d)(x) ∧ d ≥ sidiot ∧ λp[terrible(p)](ˆidiot(d)(x))]
= λx∃d[idiot(d)(x) ∧ d ≥ sidiot ∧ terrible(ˆidiot(d)(x))]
12 I am leaving the details of this aside. See Morzycki (2004) for some suggestions as to how
the individual- and proposition-modifying uses of evaluative adjectives may be related.
59
3.3
Conclusion
In this chapter, I have compared Morzycki’s (2009) analysis of size adjectival
degree modification with the previous chapter’s analysis of evaluative degree
modification involving eval and a propositional modifier. From this, DegN P
has emerged as a relatively simple structure with room for one degree head and
one modifier (in Spec):
(41)
DegN Phe,ti
AP
DegN
pos/
measN /
evalN
NPhd,eti
measN and evalN differ from their counterparts in the adjectival DegP in
incorporating the denotation of pos.
Moreover, in a slightly more speculative argument, I have established a threeway distinction between concepts denoted by nouns. The class of concepts that
corresponds to the class of nouns that is linguistically gradable is a) conceptually
gradable, i.e. has fuzzy category membership, and b) lacks a prototype that
embodies the concept’s dimensions more than any other member of the category.
The linguistic gradability of this class (which contains nouns like idiot, nerd,
fan, psychopath, enthusiast, and weirdo) was demonstrated by showing that
they are subject to monotonicity-related phenomena.
60
Chapter 4
Suggestions and further
issues
In this final chapter, I would like to open discussion on two issues that came up
while I was writing this thesis.
In section 4.1, I will present data from several languages that suggest that
the class of evaluative adjectives that cannot modify degree in nouns is strikingly similar across languages, which shows that an analysis of evaluative degree
modification might not be entirely complete without a deeper investigation into
the lexical semantics of evaluatives.
Secondly, in section 4.2 I will return to verbs like appreciate, briefly mentioned in the introduction, and argue that there are several reasons to assume
that the class of gradable predicates includes verbs as well.
4.1
Evaluative modif iers across languages
In English, all evaluative adverbs seem to be able to function as degree modifiers;
the class of evaluative adjectives that can do so, however, is much smaller.
Compare for example the sentences in (42) with those in (43):
(42)
(43)
a. Maxwell is unbelievably nerdy.
b. Maxwell is an unbelievable nerd.


 surprisingly 
unexpectedly
a. Maxwell is
nerdy.


abnormally


 surprising 
unexpected
b. %Maxwell is a(n)
nerd.


abnormal
The degree reading is present in (43a), but, puzzlingly, absent in (43b). The
class of adjectives that behave like unbelievable - i.e. result in a degree reading 61
includes unbelievable, unimaginable, extraordinary, and, depending on the context1 , terrible, amazing, and insane, among others. The class of adjectives that
behave like surprising includes unexpected, bizarre, unacceptable, and abnormal.
Why should this difference exist? There are two possible hypotheses: the
first is that we are dealing with some historical coincidence that makes it possible for some adjectives to modify degree, but not for others. The difference
between unbelievable and surprising would then simply be a quirk of English.
The second hypothesis is that there is some inherent semantic difference between unbelievable- and surprising-type adjectives that lies at the root of their
behaviour as modifiers. In this case, we would expect similar adjectives in other
languages to pattern with their English counterparts. Of course, a combination
of the two hypotheses is possible as well: for some adjectives, their behaviour
could be idiosyncratic, while for others it could reflect some fundamental semantic property and hence hold crosslinguistically.
In the following section, I present data from eight different languages that
show that, while the behaviour of many evaluative modifiers seems indeed
language-specific, the crosslinguistic equivalents of surprising and unexpected
are, almost without exception, unable to modify degree. While more research
is obviously necessary (nine languages is hardly a representative sample), this
strongly suggests that the ability of these adjectives to modify degree should,
somehow, be sought in their semantics.
4.1.1
Data
The data presented in this section were elicited by means of a questionnaire,
which is included as Appendix A (the exceptions are Norwegian and Swedish,
for which I used a simpler questionnaire, asking only after a small set of specific
evaluatives). With the exception of Dutch, the data for each language were
provided by a single informant. For each language, two examples are included:
one in which the construction has a degree reading, and one in which it has not.
(44)
a. Degree reading:
(Dutch)
Kees-Jan is een onvoorstelbaar genie.
‘Kees-Jan is an unimaginable genius’
b. No degree reading:
Kees-Jan is een verrassende dikkerd.
‘Kees-Jan is a surprising fat.person’
c. Other modifiers that result in a degree reading: ongelofelijk ‘unbelievable’, onvoorstelbaar ‘unimaginable’, vreselijk /verschrikkelijk
‘terrible’
d. Other modifiers that fail to result in a degree reading: onverwacht
‘unexpected’, idioot ‘insane’, abnormaal ‘abnormal’
1 Cf.
section 4.1.2
62
(45)
a. Degree reading:
Justin är en otrolig knäppis.
‘Justin is an unbelievable weird.person’
(Swedish)
b. No degree reading:
Justin är en överraskande knäppis.
‘Justin is a surprising weird.person’
(46)
a. Degree reading:
(Norwegian)
Justin er en fryktelig raring.
‘Justin is a terrible weirdo’
b. No degree reading:
Justin er en overraskende raring.
‘Justin is a surprising weirdo’
(47)
a. Degree reading:
(French)
Cette pice fut un incroyable succès.
‘This play was an incredible success’
b. No degree reading:
Marie est une passione de tricot anormale.2
Marie is an enthusiast of knitting abnormal
’Marie is an abnormal knitting enthusiast’
c. Other modifiers that result in a degree reading: fou/folle ‘insane’,
poufantable ‘dreadful’, stupfiant ‘amazing’, surprenant ‘surprising’,
impossible, incompréhensible, inimaginable, extra-ordinaire, remarquable
d. Other modifiers that fail to result in a degree reading: inattendu
‘unexpected’, effrayant ‘frightening’, horrible, bizarre, abnormal
(48)
a. Degree reading:
(Slovene)
Jaka je neverjeten idiot.
Jake is unbelievable idiot
‘Jake is an unbelievable idiot’
b. No degree reading:
Jaka je presenetljiv/nepričakovan idiot.
Jake is surprising/unexpected idiot
‘Jake is a surprising/unexpected idiot’
c. Other modifiers that result in a degree reading: neverjeten ‘unbelievable’, izjemen ‘exceptional’, strašen ‘horrific’
(49)
a. Degree reading:
(Mandarin)
2 But note that the Position Generalisation predicts that a degree reading would not be
possible at all with this word order, regardless of the evaluative modifier used.
63
John shi4 ge nan2 yi3 xiang3 xiang4 de ben4 dan4 .
John is CL unimaginable idiot
‘John is an unimaginable idiot’ (degree reading)
b. No degree reading:
John shi4 ge shi1 qu4 li3 zhi4 de ben4 dan4 .
John is CL insane idiot
‘John is an idiot that he lost his mind’
c. Other modifiers that result in a degree reading: nan2 yin3 zhi4 xin4 de
‘unbelievable’, nan2 yi3 xiang3 xiang4 de ‘unimaginable’, bu4 ke3 si1 yi4 de
‘incredible’, wu2 fa3 li2 jie3 de ‘incomprehensible’
d. Other modifiers that fail to result in a degree reading: bu2 zheng4 chang2 de
‘abnormal’, fei1 fan2 de ‘extraordinary/remarkable’, ling4 ren2 jing1 ya4 de
‘surprising’, wu2 fa3 yu4 liao4 de ‘unexpected’, ke3 pa4 de ‘horrible’, emphkong3 bu4 de
‘dreadful’, ling4 ren2 jing1 qi2 de ‘amazing’
(50)
a. Degree reading:
(Japanese)
Tom-wa namihazureta tensai desu.
Tom-TOP extraordinary genius is
‘Tom is an extraordinary genius’
b. No degree reading:
Tom-wa kikaina yakyuu aikooka desu.
Tom-TOP bizarre baseball fan is
‘Tom is a bizarre baseball fan.’
c. Other modifiers that result in a degree reading: odorokubeki ‘amazing’
(51)
a. Degree reading:
(Hebrew)
hu idiot yotze dofen
he idiot out.of frame
‘He’s an extraordinary idiot’
b. Other modifiers that result in a degree reading: ‘impossible’, ‘extraordinary’ (∼‘out of frame’)3
c. Other modifiers that fail to result in a degree reading: ‘surprising’,
‘unexpected’ (∼‘not predictable’), ‘dreadful’, ‘frightening’, ‘amazing’
4.1.2
Some tentative conclusions
Positive and negative meanings. As can be seen in the data above, adjectives whose meaning is either very positive or very negative (like amazing,
horrible or dreadful ) show very mixed behaviour across languages. In Dutch,
for example, horrible-like adjectives can quite easily modify degree, losing their
3 My informant provided these adjectives only in Hebrew script, so I will give just the
translation.
64
negative connotation (for example, it is perfectly possible to call oneself a verschrikkelijke balletliefhebber ‘horrible ballet-lover’) without implying that there
is anything particularly horrible about loving ballet. The opposite picture
emerges in Mandarin Chinese, in which, according to my informant, adjectives
like horrible and dreadful cannot modify degree in nouns because “they cannot
be used metaphorically” - their use always implies a negative judgement, which
apparently blocks a degree reading somehow. However, they are all perfectly
able to modify degree in adverbial form, in which they turn into ‘metaphors’
for very high degrees, just as in Dutch.
Similar observations are made by Paradis (2000), who argues that terrible
and its kin have undergone a historical process from propositional modifiers to
reinforcers, “[t]hat is, they have lost their propositional content and become
exponents of an abstract notion of degree. (...) The reinforcing reading (...)
is characterized by a conceptualization where the gradable mode of construal
is in the foreground almost to the exclusion of the content domain. (...) [The
adjectives] have developed into unbounded markers of HIGH DEGREE.”
In other words, there seems to be a crosslinguistic correlation between the
availability of a degree reading and the possibility to interpret the adjective
neutrally. This may even vary between individuals: in spite of Paradis’s observations about adjectives like terrible, some of the native English speakers I
consulted were unable to get a degree reading for terrible nerd because they
“regarded being nerdy as a positive thing”.
Is this correlation really there, and if so, why would it exist? Why does it
matter with gradable nouns, but not with adjectives?
Surprising and unexpected. While other specific evaluatives do not really
show a cross-linguistic pattern, surprising and unexpected seem to be generally
unable to modify degree (one exception is French surprenant; for Japanese, I
sadly do not have the relevant data). Although the sample is tiny and the
observation might, theoretically, be due to chance, I feel it is at least suggestive
enough to merit further investigation.
At the moment, I do not have an explanation of what would make surprising
and unexpected different from other evaluatives. They seem more readily gradable; a Forestle4 search makes clear that 8.9% of the occurrences of surprising
and 2.4% of the occurrences of amazing are graded by either very, extremely,
rather, surprisingly, a bit, or incredibly, against on average 0.9% of the occurrences of unbelievable, incredible, unimaginable, impossible and extraordinary.
But this might be a consequence rather than a cause: as was suggested above,
the ability to modify degree seems to go hand in hand with a certain neutral,
‘abstract’ interpretation, and modifying degree in such an abstract modifier may
quickly get too complicated or costly to interpret.
The above observations suggest that the lexical semantics of evaluatives is important to their ability to modify degree in several ways. However, they also
4A
Yahoo-based search engine that saves a tiny patch of rainforest with every search.
65
touch upon the much broader issue of the difference between adverbs and adjectives: while many evaluative adjectives, across languages, are unable to modify
degree in nouns, we have yet to see an evaluative adverb that cannot modify
degree. Could the fact that DegN P seems to be much less sophisticated than
its adjectival equivalent play some role in drawing out small lexical distinctions
between different evaluatives? If so, how?
4.2
Gradable verbs
In the introduction (section 1.1), we briefly considered the following sentence:
(52)
Maxwell appreciates jazz more than Vernon.
There clearly seems to be something ‘gradable’ about the verb appreciate, in
the sense that it is a predicate that can apply to an individual to a smaller or
larger extent. But is it actually gradable in the sense relevant here, i.e. that of
having a degree argument?
According to Kennedy & McNally (2005), the degree scale structure of deverbal adjectives (much appreciated, very surprised, well understood ) derives
from certain semantic properties of the verb; its event structure determines the
adjective’s scale structure, and the various ways in which appreciation, surprise
and understanding can be measured determine the scale’s dimension:
For example, arguably the most natural reading of much admired is
paraphrasable as admired by many people, without necessarily entailing that the admiration has been long lasting; at the same time,
a much talked about program might well be one that has been talked
about many times, though not necessarily by many different people
or for a particularly long time on any given occasion. Adjectives
denoting the property of being the object of an emotion, such as
admired, loved, or hated, or the experiencer of an emotion, such as
worried, also permit readings on which the scalar dimension along
which they are measured is one of intensity. For example, a much
despised neighbor might be despised by only one individual, but with
a passion.
What these facts indicate is that the dimensional parameter of the
derived scale, like the structure of the scale, is also a function of the
meaning of the source verb. Specifically, any of the various aspects of
verb meaning that support measurement (temporal extent, number
of occurrences, number of participants, intensity, etc.) can be used
to fix the dimensional parameter of the derived adjectives scale.
In other words, according to Kennedy & McNally, the intuitive ‘gradability’ of a
verb like appreciate, which is best defined by what they call intensity, is merely
one of the dimensions across we could measure appreciation - it is determined
by the context in which we use the verb and does not indicate the presence of a
66
degree argument. This idea is very reminiscent of what I have called ‘conceptual
gradability’ in nouns - and just like before, we may ask if conceptual gradability
is really all there is to it.
I propose to consider the possibility that verbs with a degree argument what we have termed ‘linguistically gradable’ - do exist. In fact, the existence
of gradable verbs should be our null hypothesis: verbs are predicates, just like
adjectives and nouns, and as the latter two can be either gradable or nongradable, we would expect verbal predicates to show the same distinction.
Of course, the question is whether there is empirical support for our null
hypothesis. If we could show that appreciate and its kin can occur in degree
constructions that require the presence of a degree argument, this would be
evidence for the actual gradability of this class of verbs. This sounds easier
than it is. The elements that deal with a gradable predicate’s degree argument
are the degree heads we saw in Chapter 1 - so, how, too, as, pos and meas. The
first four cannot have a verbal complement, for independent syntactic reasons;
the last two are phonologically null, so demonstrating their presence is not an
easy matter. However, there is some evidence that points towards it.
Entailment of the positive form. First, for some verbs, the entailment
patterns of the verb’s ‘positive form’ and the verbal comparative are identical
to the entailments found in parallel adjectival constructions, which points to the
presence of pos. To see why this is so, consider the following data:
(53)
a. Vernon is tall.
∼Vernon’s height exceeds the standard.
b. Vernon appreciates jazz.
∼Vernon’s appreciation of jazz exceeds the standard.
(54)
a. Vernon is taller than Maxwell, but he is still pretty short.
b. Maxwell hates jazz with a vengeance. Vernon appreciates it more
than Maxwell, but he still rather dislikes it.
The contrast between (53) and (54) shows that the ‘above-standard’ implication
of appreciate is not part of the verb itself, as it is absent from the comparative
construction (if it were present, (54b) would be contradictory). This exactly
mirrors the semantics of the adjective, whose positive form is not entailed by
the comparative. So, if the verb itself does not involve this comparison with a
standard value, what other element could contribute this but pos?
Modification by size adverbs. Secondly, it was argued earlier in this chapter (following Morzycki 2009) that degree modification by size adjectives and
adverbs is possible because size adjectives are abstract enough to be able to
measure the size of degrees. In other words, a predicate cannot be modified by
a size adjective unless there is a degree to measure, as we can see by contrasting
gradable and non-gradable predicates:
(55)
a. Maxwell is hugely/enormously tall.
67
b. *Maxwell is hugely/enormously dead.
(56)
a. Maxwell is a huge/enormous Barbie doll enthusiast.
b. %Maxwell is a huge/enormous soccer player.
Crucially, the same contrast emerges when we compare appreciate with other,
clearly non-gradable verbs:
(57)
a. Maxwell hugely/enormously appreciates jazz.
b. Vernon hugely/enormously regrets his past decisions.
c. *Maxwell hugely/enormously sprayed the wall with paint.
d. *Vernon hugely/enormously rides his bicycle in the morning.
Note that the ungrammaticality of (57c,d) does not in any way imply that
their deverbal adjectives should not be gradable. They are: walls can be fully
sprayed or well sprayed or more sprayed with paint than the wall opposite it, and
bicycles can be much ridden. If apparent gradability in verbs was just a matter
of contextually associated dimensions, and hence appreciate and regret were
gradable because of the intensity dimension associated with it, why would the
same not hold for ride? After all, the corresponding deverbal adjective ridden
and the possibility of modifying it with much (Maxwell does not ride his bicycle
much) clearly shows that it has a salient number-of-occurences dimension.
In short, while the availability of size adverbial modification already points
to the presence of a degree argument, this conclusion is supported even more
by the otherwise inexplicable contrast with other, non-gradable verbs.
Monotonicity. Finally, if verbs have a degree argument, we expect them to be
monotone just like other gradable predicates. Recall that monotonicity explains
why (58a) is not true of a situation in which Vernon is tiny or has a height equal
to his birthday, and why (58b) and (58c) lack a degree reading:
(58)
a. Vernon is unusually tall.
b. %Vernon is usually tall.
c. %Vernon is a diminutive jazz lover.
Indeed, the same seems to hold for verbs:
(59)
a. Vernon appreciates jazz extraordinarily.
appreciates jazz
b. %Vernon
ordinarily.
regrets his decision
appreciates jazz
c. %*Vernon
diminutively.
regrets his decision
We cannot use (59a) to indicate that Vernon takes his dislike of jazz to unhealthy levels, or (59b) to state that he appreciates jazz about as much as the
next guy. Finally, (59c) shows that the Bigness Generalisation - which was
68
explained in terms of monotonicity - applies to size adverbial modification of
verbs as well: hugely and enormously can easily modify degree in verbs, but
diminutively cannot.
Concludingly, I would like to suggest that verbal gradability is not just a matter
of conceptual, but also of linguistic gradability. The issue of verbal gradability
is obviously much more complicated than I have shown, or could have shown,
in a thesis whose scope is adjectives and nouns, but I hope this brief discussion
has at least opened a new perspective on the matter. Much more needs to be
said, I think, about the ways gradable predicates from different lexical classes
are semantically interwoven.
69
Chapter 5
Conclusions
In this thesis, I have argued for an analysis of evaluative degree constructions
that - rather than analysing them as embedded exclamatives - treats them
largely on a parallel with other, more widely studied degree constructions. This
analysis explains a wide range of linguistic phenomena: it correctly predicts
the absence of much-support with EDCs in English, and the behaviour of gender agreement in Dutch. Moreover, it allows EDCs to be headed by different
degree morphemes (all except for meas, which is ruled out for independent reasons), and - because EDCs follow the standard recipe for a degree construction:
a degree predicate, a comparison function and a comparison value - correctly
predicts a semantics for each of these combinations that parallels their unmodified counterparts. The analysis also explains, in a straightforward way, the
intuitive and empirical differences and similarities between evaluative modifiers
and ‘pure’ degree modifiers like very, by assigning them to the same syntactic
position, but a different semantic type.
Another debate that has returned several times throughout this thesis concerns the analysis of gradable predicates as either degree predicates (type hd, eti)
or measure functions (type he, di). I have argued that evaluative degree modification, as far as I can tell, cannot be explained under a measure function-based
approach; in any case, a measure function-based analysis of EDCs will bear very
little resemblance to the one presented in this thesis, and presumably lack its
empirical benefits. When it comes to size adjectival modification of gradable
nouns, the two approaches are equal in explanatory power, although the measure function approach needs a more heavy adjustment of the semantics of meas
compared to its adjectival counterpart than the degree predicate approach does.
Within the degree predicate approach, the notion of monotonicity is crucial.
It explains the absence of certain interpretations of EDCs, and the inability of
certain modifiers - upward monotone evaluatives and size adjectives predicating smallness - to modify degree. Because monotonicity is a property of degree
predicates, it can also be used as a test to see whether a seemingly gradable
predicate does in fact have a degree argument. This way, I have shown certain
nouns and verbs to be linguistically gradable, just like adjectives.
70
A secondary aim of this thesis has been to flesh out the notion of ‘gradable
noun’. I have argued that a distinction should be made between conceptual and
linguistic gradability, the latter being defined by the presence of a degree argument. The existence of a phenomenon like metalinguistic comparison, which
is sensitive to conceptual gradability, shows that apparent degree modification
in nominal constructions does not necessarily point to the presence of a degree
argument. By testing for the presence or absence of degree readings with modifiers that are sensitive to the predicate’s monotonicity, however, it is possible to
show that there is a class of nouns that is linguistically gradable. I have defined
this class in prototype-theoretic terms as those nouns that denote concepts that
a) have graded membership, and b) lack a prototype.
Suggestions for further research have included the exploration of DegP structure
for gradable mass nouns and verbs, and a closer look into the lexical semantics
of evaluatives to explain some differences in their behaviour. It was suggested,
based on crosslinguistic data, that the ability of evaluative adjectives to modify
degree in nouns is correlated with the possibility of ‘neutralising’ their evaluative
content. How this ties in with the semantics for EDCs presented in this thesis
(which, in principle, keeps the evaluative content of the modifiers intact, treating
them like propositional modifiers) remains a topic for further investigation.
71
References
Abney, Steven (1987). The English Noun Phrase in its Sentential Aspect. PhD
dissertation, MIT, Cambridge, MA.
Armstrong, Sharon, Lila Gleitman & Henry Gleitman (1983). ‘What some
concepts might not be’. Cognition 13.
Bierwisch, Manfred (1989). ‘The semantics of gradation’. In: Bierwisch, Manfred & Ewald Lang (eds). Grammatical Structure and Conceptual Interpretation. Berlin: Springer-Verlag.
Bolinger, Dwight (1972). Degree Words. The Hague: Mouton.
Bresnan, Joan (1973). ‘Syntax of the comparative clause construction in English’. Linguistic Inquiry 4:3, 275-434.
Chomsky, Noam (1986). Barriers. Cambridge, MA: MIT Press.
Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: MIT
Press.
Chomsky, Noam (2004). ‘Beyond Explanatory Adequacy’. In: Belletti, Adriana (ed.). Structures and Beyond. New York: Oxford University Press.
Cinque, Guglielmo (1994). ‘On the evidence for partial N-movement in the
Romance DP’. In: Cinque, Guglielmo, Jan Koster, Jean-Yves Pollock,
Luigi Rizzi & Raffaella Zanuttini (eds). Paths Toward Universal Grammar, 85110. Georgetown: Georgetown University Press.
Corver, Norbert (1991). ‘Evidence for DegP’. Proceedings of NELS 21, Umass,
Amherst.
Corver, Norbert (1997a). ‘Much-support as a last resort’. Linguistic Inquiry
28:1, 119-164.
Corver, Norbert (1997b). ‘The external syntax of the Dutch adjectival projection’. Natural Language and Linguistic Theory 15, 289-368.
Corver, Norbert (2006). ‘Proleptic agreement as a good design property’. In:
Costa, João & Maria Cristina Figueiredo. Studies on Agreement. Amsterdam: John Benjamins.
72
Corver, Norbert (2009). ‘Getting the (syntactic) measure of Measure Phrases’.
The Linguistic Review 26:1, 67-134.
Cresswell, Max (1976). ‘The semantics of degree’. In: Barbara H. Partee (ed.).
Montague Grammar, 261-292. New York: Academic Press.
Cruse, D. Alan (1980). ‘Antonyms and gradable complementaries’. In: Kastovsky, Dieter (ed.). Perspektiven der Lexikalischen Semantik, 1425. Bonn:
Bouvier.
Doetjes, Jenny (1997). Quantifiers and Selection. PhD dissertation, Leiden
University.
Doetjes, Jenny (2008). ‘Adjectives and degree modification’. In: McNally,
Louise & Christopher Kennedy (eds). Adjectives and Adverbs: Syntax,
Semantics, and Discourse. Oxford: Oxford University Press.
Fukui, Naoki & Margaret Speas (1986). ‘Specifiers and projection’. MIT
Working Papers in Linguistics 8.
Giannakidou, Anastasia & Melita Stavrou (2009). ‘On metalinguistic comparatives and negation in Greek’. MIT Working Papers in Linguistics
57.
Giannakidou, Anastasia & Suwon Yoon (2009). ‘Metalinguistic comparatives
in Greek and Korean: Attitude semantics and expressive content’. In: Riester, Arndt & Torgrim Solstad (eds). Proceedings of Sinn und Bedeutung
13, University of Stuttgart.
Heim, Irene & Angelika Kratzer (1998). Semantics in Generative Grammar.
Oxford: Blackwell.
Heim, Irene (2000). ‘Degree operators and scope’. In: Jackson, Brenda &
Tanya Matthews (eds). Proceedings of Semantics and Linguistic Theory
(SALT) 10. Ithaca, NY: CLC Publications.
Hornstein, Norbert, Jairo Nunes & Kleanthes Grohmann (2005). Understanding Minimalism. Cambridge: Cambridge University Press.
Jackendoff, Ray (1977). X-Bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press.
Kamp, Hans & Barbara Partee (1995). ‘Prototype theory and compositionality’. Cognition 57, 129-91.
Karttunen, Lauri (1977). ‘Syntax and semantics of questions’. Linguistics and
Philosophy 1:1, 3-44.
Katz, Graham (2005). ‘Attitudes towards degrees’. In: Maier, Emar, Corien
Bary & Janneke Huitink (eds). Proceedings of Sinn und Bedeutung 9.
73
Kennedy, Christopher & Louise McNally (2005). ‘Scale structure, degree modification, and the semantics of gradable predicates’. Language 81:2, 345381.
Kennedy, Christopher (1997). Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. PhD dissertation, University of
California, Santa Cruz.
Kennedy, Christopher (2001). ‘Polar opposition and the ontology of ‘degrees”.
Linguistics and Philosophy 24, 33-70.
Kennedy, Chrisopher (2004). ‘Comparatives, Semantics of’. In: Allen, Keith
(ed.). Encyclopedia of Language and Linguistics, Second Edition, Lexical
and Logical Semantics section. Oxford: Elsevier.
Klein, Ewan (1980). ‘A semantics for positive and comparative adjectives’.
Linguistics and Philosophy 4, 1-45.
Liu, Chen-Sheng Luther (in press). ‘The positive morpheme in Chinese and
the adjectival structure’. Lingua, doi:10.1016/j.lingua.2009.06.001
McCawley, James (1988). The Syntactic Phenomena of English. Chicago:
University of Chicago Press.
Morzycki, Marcin (2004). ‘Evaluative adverbial modification in the adjectival
projection’. Ms, Université du Québec à Montréal.
Morzycki, Marcin (2008). ‘Metalinguistic comparison in an alternative semantics for imprecision’. In: Abdurrahman, Muhammad, Anisa Schardl, &
Martin Walkow (eds). Proceedings of NELS 38. Amherst, MA: GLSA
Publications.
Morzycki, Marcin (2009). ‘Degree modification of gradable nouns: size adjectives and adnominal degree morphemes’. Natural Language Semantics 17,
175-203.
Nicolas, David (2004). ‘The semantics of nouns derived from gradable adjectives’. In: Meier, Cécile, & Matthias Weisgerber (eds). Proceedings of
Sinn und Bedeutung 8, Goethe University Frankfurt.
Nouwen, Rick (2005). ‘Monotone amazement’. In: Dekker, Paul & Michael
Franke (eds). Proceedings of the Fifteenth Amsterdam Colloquium, University of Amsterdam.
Nouwen, Rick (2009). ‘Degree modifiers and monotonicity’. Ms, Utrecht institute of Linguistics OTS.
Paradis, Carita (2000). ‘Reinforcing adjectives: a cognitive semantic perspective on grammaticalization’. In: Berm’udez-Otero, Ricardo, David Denison, Richard M. Hogg, C.B. McCully (eds). Generative theory and corpus studies. Topics in English Linguistics, 233-258. Berlin: Mouton de
Gruyter.
74
Rapp, Irene & Arnim von Stechow (1999). ‘Fast ‘almost’ and the visibility
parameter for functional adverbs’. Journal of Semantics 16:2, 149-204.
Rett, Jessica (2008). Degree Modification in Natural Language. PhD dissertation, Rutgers, The State University of New Jersey.
van Rooij, Robert (to appear). ‘Vagueness and Linguistics’. In: Ronzitti,
Giuseppina (ed.). The Vagueness Handbook. Dordrecht: Springer.
Rosch, Eleanor (1973). ‘On the internal structure of perceptual and semantic categories’. In: Moore, T.E. (ed.). Cognitive Development and the
Acquisition of Language. New York: Academic Press.
Rosch, Eleanor (1978). ‘Principles Of Categorization’. In: Rosch, Eleanor &
B.B. Lloyd (eds). Cognition and Categorization. N.J.Erlbaum, Hillsdale.
Rotstein, Carmen & Yoad Winter (2004). ‘Total adjectives vs. partial adjectives: scale structure and higher-order modifiers’. Natural Language
Semantics 12, 259288.
Sassoon, Galit Weidman (2007). Vagueness, Gradability and Typicality: A
Comprehensive Semantic Analysis. PhD dissertation, Tel Aviv University.
Schoorlemmer, Erik (2009). Agreement, Dominance and Doubling: The Morphosyntax of DP. PhD dissertation, Leiden University.
Schwarzschild, Roger (2005). ‘Measure phrases as modifiers of adjectives’. In:
L’Adjectif, Recherches Linguistiques de Vincennes 35, 207-228.
Svenonius, Peter & Christopher Kennedy (2006). ‘Northern Norwegian degree
questions and the syntax of measurement’. In: Frascarelli, Mara (ed.).
Phases of Interpretation, 133-161. Berlin: Mouton de Gruyter.
Sybesma, Rint (1999). The Mandarin VP. Dordrecht: Kluwer Academic Publishers.
Syrett, Kristen & Jeffrey Lidz (in press). ‘30-month-olds use the distribution
and meaning of adverbs to interpret novel adjectives’. Language Learning
and Development.
von Stechow, Arnim (1984). ‘Comparing semantic theories of comparison’.
Journal of Semantics 3, 1-77.
Winter, Yoad (2009). ‘What neutralizes the positive?’ Talk presented at the
‘Degrees under Discussion’ mini-workshop, Utrecht institute of Linguistics
OTS. http://www.phil.uu.nl/∼yoad/papers/WinterAdjectivesTalk.pdf
Wittgenstein, Ludwig (1953/2001). Philosophical Investigations. Translated
by G.E.M. Anscombe. Third edition. Oxford: Blackwell.
75
Yoon, Youngeun (1996). ‘Total and partial predicates and the weak and strong
interpretations’. Natural Language Semantics 4, 217-236.
Zamparelli, Roberto (1993). ‘Pre-nominal modifiers, degree phrases and the
structure of AP’. University of Venice Working Papers in Linguistics 3:1.
Zanuttini, Raffaella and Paul Portner (2003). ‘Exclamative clauses: at the
syntax-semantics interface’. Language 79:1.
Zwicky, Arnold (1970). ‘Usually and unusually’. Linguistic Inquiry 1:1.
76
Appendices
77
Appendix A: Questionnaire
Background
Every languages contains adjectives that can be modified to indicate the degree to which they hold. For example, we can modify tall in John is tall to
indicate how big John is: rather tall, very tall, extremely tall, ridiculously tall,
and so on. These gradeable adjectives have been widely studied by syntacticians
and semanticists.
But adjectives are not the only words that can be graded. There are also
gradeable nouns, although this is a much less studied phenomenon. Examples
are nouns like nerd, idiot, fan, enthusiast, simpleton, and so on. Obviously, one
can be an idiot or knitting enthusiast to a greater or lesser degree, and just as
with the adjectives this degree can be indicated by adding a modifier. A wellknown example of such a modifier is big (and similar adjectives): we interpret
John is a big/huge/enormous knitting enthusiast as ’John is very enthusiastic
about knitting’, not as ’John is a knitting enthusiast and physically big’.
However, the class of possible noun modifiers appears to be much smaller than
the class of possible adjective modifiers. Compare the following sentences:
(a) John is abnormally nerdy
(b) John is an abnormal nerd
(c) John is surprisingly idiotic
(d) John is a surprising idiot
While abnormally in (a) has a degree reading ’John is nerdy to an abnormal
degree’ meaning ’John is extremely nerdy’ abnormal in (b) hasn’t; (b) can only
mean (at least to the overwhelming majority of English speakers I consulted)
that John is a nerd and also abnormal (or that John is a nerd but not in a ’normal’ way). Similarly, while (c) means ’John is idiotic to a surprising degree’, (d)
doesn’t: it only has the reading ’John is an idiot and also surprising’ (whatever
that may mean).
78
One of the purposes of my thesis is to find out which adjectives can function as degree modifiers. Basically, there are two options: (1) something in
the semantics of certain adjectives makes them appropriate as degree modifiers,
and (2) degree-modifying adjectives simply have this function as part of their
lexicalised meaning, and the difference between adjectives that can and can’t
modify a gradeable noun is basically a historical coincidence.
In the second case, we’d expect the class of possible modifiers to be very
different across languages, whereas in the first case we’d expect to see similar properties for words with a similar semantics crosslinguistically. And this,
obviously, is where you speaker of an exotic language (i.e. not English) come in!
I know it’s a lot, but I’d be extremely happy if you could help me. Even if
you can provide only a few adjectives, it’s better than nothing. Thanks a lot!
Not sure how to answer a certain question? An example of a completed questionnaire (for Dutch) is also included.
Questionnaire
Instructions for question 2 and 3: The general format to test for degree
readings is as follows:
• A is your favourite gradeable noun (a direct translation of one of the English ones mentioned above, or a different one).
• B is an adjective.
• Using your language’s equivalent of John is a B A, try to decide which
interpretation fits this sentence best:
(a) the degree reading: ’John is A to a B degree’ / ’John is very A-like’
(b) the non-degree reading, approximated by ’John is an A, and he is
also B’ or something like ’John is an A in a B sort of way’
(1) Which language do you speak?
(2) Can your gradeable noun be modified by your language’s equivalent of big,
and does this result in a degree reading? Can you give an example?
(3) Of the list below, which adjectives result in a degree reading? Which do
not? Can you provide:
– a list of adjectives with English translation, as literal as possible, plus
an indication of whether they have a degree reading or not (if you’re
not sure, or if it depends on the context, just write that down)
79
– one or two complete examples (the adjective in a John is a B A
sentence, with glosses (a very literal word-for-word translation into
English))
I am especially interested in your language’s equivalents of the following English
adverbs:
• unbelievable, incredible
• surprising, unexpected
But also (if you have enough time):
• incomprehensible, unimaginable, impossible, abnormal, insane, bizarre,
extraordinary, remarkable, horrible, dreadful, frightening, amazing...
Example questionnaire
(1) Dutch.
(2) Yes; example: ’Kees-Jan is een groot fan van klassiek ballet’
Kees-Jan is a big fan of classical ballet
Interpretation: K-J really likes classical ballet
(3) Adjectives that result in a degree reading: ongelofelijk (’unbelievable’),
onvoorstelbaar (’unimaginable’), ongekend (lit. ’unknown’: ’unheard-of’),
ongehoord (’unheard-of’), ontzettend (lit. ’horrible, frightful’, though not
often used in this interpretation anymore), onmogelijk (’impossible’), extreem (’extreme’)
Example: ’Kees-Jan is een onvoorstelbaar genie’
Kees-Jan is an unimaginable genius
Interpretation: K-J is brilliant to an unimaginable degree
(≈ K-J is very brilliant)
Adjectives that do not result in a degree reading: verrassend (’surprising’),
bizar (’bizarre’), buitengewoon (’extraordinary’), abnormaal (’abnormal’),
onverwacht (’unexpected’), idioot (’insane’), opmerkelijk (’remarkable’)
Example: ’Kees-Jan is een buitengewone dikkerd’
Kees-Jan is an extraordinary fat.person
Interpretation: K-J is an extraordinary guy who is also fat
80